Tree-Based vs. SNP-Based Introgression Tests: A Comprehensive Performance Review for Genomic Research

Abstract

This article provides a systematic comparison of tree-based and SNP-based methodologies for detecting introgression, a key evolutionary process with significant implications for adaptation and disease research. Aimed at researchers and biomedical professionals, we explore the foundational principles of both approaches, detail their practical application using modern software tools, and address critical troubleshooting scenarios, including false positives caused by rate variation and homoplasy. Through a validation framework incorporating simulation studies and real-world genomic case studies, we deliver evidence-based recommendations for method selection to enhance accuracy in evolutionary genomics and the identification of adaptively introgressed loci in biomedical research.

Understanding Introgression Detection: Core Principles of Phylogenetic and SNP-Based Methods

Defining Introgression and Its Impact on Evolution and Adaptation

Introgression, also known as introgressive hybridization, is the transfer of genetic material from one species into the gene pool of another through the repeated backcrossing of an interspecific hybrid with one of its parent species [1]. This evolutionary process differs from simple hybridization, which results in a relatively even mixture of parental genes in the first generation, as introgression produces a complex, highly variable mixture that may involve only a minimal percentage of the donor genome [1]. Over the past decade, advances in genomic technologies have transformed our understanding of introgression, revealing it to be a widespread phenomenon across the tree of life with significant implications for adaptation, speciation, and conservation biology [2]. This review examines the defining characteristics of introgression and compares the performance of different methodological approaches for its detection, with particular emphasis on tree-based versus SNP-based tests within the context of contemporary genomic research.

What is Introgression?

Introgression represents a long-term evolutionary process that requires multiple generations of backcrossing before significant incorporation of foreign genetic material occurs [1]. The process begins when matings between members of two species produce partially viable and fertile hybrid offspring, which then reproduce with members of one or both parental species [2]. Through successive generations of backcrossing, DNA from one species becomes permanently incorporated into the genome of another [2]. This process is distinct from incomplete lineage sorting, which can produce similar genetic patterns but occurs due to deep ancestral genetic variation rather than secondary genetic exchange [2].

A particularly important evolutionary dimension is adaptive introgression, which occurs when the incorporation of foreign genetic variants increases the overall fitness of the recipient population [1] [3]. Unlike neutral introgression, which may be lost through genetic drift, adaptive introgression is maintained by natural selection and can lead to the rapid fixation of beneficial alleles [3]. Research across diverse taxonomic groups has demonstrated that adaptive introgression can facilitate evolutionary leaps by bypassing intermediate evolutionary stages, allowing species to respond more quickly to environmental changes than would be possible through de novo mutations alone [3].

The Evolutionary Impact of Introgression

Adaptive Advantages

Introgression serves as a significant source of genetic variation in natural populations and can contribute substantially to adaptation and even adaptive radiation [1]. By introducing genetic variation that has been "pre-tested" by selection in another species, introgression allows populations to evolve rapidly in response to environmental challenges [2]. Documented cases of adaptive introgression span diverse organisms:

• Humans: Modern human populations carry introgressed genes from Neanderthals and Denisovans that provided adaptive advantages in immune function and high-altitude adaptation [1] [2].
• Butterflies: In Heliconius butterflies, introgression of wing pattern loci has facilitated mimicry complexes that provide protection from predators [1] [2].
• Plants: Sunflowers and Arabidopsis have acquired alleles for early flowering time and soil tolerance through introgression from closely related species [2].
• Crops: Domesticated species often benefit from introgression with wild relatives, gaining traits that enable expansion into different environments [1] [4].

Conservation Implications

While introgression can introduce beneficial genetic variation, it also poses conservation challenges, particularly when human activities alter species distributions and increase hybridization rates [2]. Genetic swamping—where hybridization and introgression drive genetic replacement of original inhabitants—becomes a concern when resident species are outnumbered by new arrivals [2]. This is particularly problematic for endangered species and locally adapted populations, as documented in European honeybees where commercial strains threaten the genetic integrity of the native Apis mellifera mellifera [5].

Comparative Performance of Introgression Detection Methods

The accurate detection of introgression represents a significant methodological challenge in evolutionary genomics. Two primary classes of statistical approaches have emerged: tree-based methods that account for evolutionary relatedness among individuals, and SNP-based methods that typically ignore these relationships [6]. The performance characteristics of these approaches differ substantially in terms of statistical power, type I error control, and applicability to different research contexts.

Methodological Frameworks

Table 1: Comparison of Introgression Detection Methods

Method Characteristic	Tree-Based Approaches	SNP-Based Approaches
Theoretical Foundation	Incorporates phylogenetic relationships and shared evolutionary history [6]	Assumes independence among observations; groups samples by allele type [6]
Detection Power	Improved ability to detect weaker associations by leveraging correlation structure [6]	Effective for detecting strong associations but may miss weaker signals [6]
Type I Error Control	Conservative error rates (below 0.05 in simulation studies) [6]	Elevated error rates (above 0.05 in some scenarios) [6]
Computational Demand	Higher due to phylogenetic tree estimation and more complex models [6]	Generally lower computational requirements [6]
Data Flexibility	Limited in handling complex covariates and biologically realistic data [6]	More flexible for incorporating external covariates and diverse data types [6]
Localization Accuracy	Better at identifying causal regions when evolutionary history is informative [6]	Preferable when association mapping is primary goal [6]

Experimental Protocols and Performance Metrics

A systematic comparison of tree-based and non-tree-based methods was conducted using simulated phenotypes on 1,943 unrelated individuals from the Genetics Analysis Workshop 19 [6]. Researchers analyzed five genes (TNN, LEPR, GSN, TCIRG1, and FLT3) with varying effect sizes using two approaches:

• Tree-Based Method: The Likelihood Score Statistic (LSS) approach estimated phylogenetic trees from SNP data, modeling trait values with a multivariate normal distribution where covariance among observations was proportional to their shared evolutionary history [6].

• SNP-Based Method: The classical 2-sample t-test grouped chromosomal observations by SNP state (minor or major allele) and performed pooled t-tests assuming independence among observations [6].

Table 2: Performance Comparison Across Gene Types

Gene (Effect Size)	Tree-Based LSS Power	SNP-Based t-test Power	Tree-Based Type I Error	SNP-Based Type I Error
TNN (Large: 10.89)	High detection power [6]	High detection power [6]	0.010 [6]	>0.05 [6]
LEPR (Large: 11.99)	High detection power [6]	High detection power [6]	0.045 [6]	>0.05 [6]
FLT3 (Medium: 3.89)	Lower power for weaker signals [6]	Lower power for weaker signals [6]	0.020 [6]	>0.05 [6]
TCIRG1 (Medium: 3.38)	Similar performance between methods [6]	Similar performance between methods [6]	0.015 [6]	>0.05 [6]
GSN (Small: 0.76)	Neither method uniquely superior [6]	Neither method uniquely superior [6]	0.020 [6]	>0.05 [6]

Advanced Detection Frameworks

More specialized methods have been developed specifically for detecting adaptive introgression. A recent performance evaluation compared three such approaches—VolcanoFinder, Genomatnn, and MaLAdapt—alongside the standalone summary statistic Q95(w, y) [7]. This study utilized simulated datasets under various evolutionary scenarios inspired by human, wall lizard (Podarcis), and bear (Ursus) lineages to represent different combinations of divergence and migration times [7]. Key findings included:

• Methods based on Q95 showed the highest efficiency for exploratory studies of adaptive introgression [7].
• The hitchhiking effect of adaptively introgressed mutations strongly impacts flanking regions, necessitating the inclusion of adjacent windows in training data for accurate identification of the target region [7].
• Performance varied significantly with divergence times, migration rates, population sizes, selection coefficients, and the presence of recombination hotspots [7].

Visualization of Methodological Approaches

The following workflow diagram illustrates the key decision points and analytical pathways for introgression detection methods:

Research Reagent Solutions for Introgression Studies

Table 3: Essential Research Tools and Resources

Research Reagent	Primary Function	Application Context
MassARRAY iPLEX System [5]	High-throughput SNP genotyping	C-lineage introgression detection in honeybees; cost-effective alternative to WGS
Customized SNP Panels [5]	Ancestry-informative marker sets	Specific introgression detection (e.g., 117-SNP panel for A. m. mellifera)
Hidden Markov Models (HMMs) [2]	Local ancestry inference	Identifying introgressed genomic segments based on spatial arrangement of sites
Conditional Random Fields (CRFs) [2]	Local ancestry inference	Alternative to HMMs for inferring probability of introgression in genomic regions
Whole-Genome Sequencing [2]	Comprehensive variant detection	Gold standard for introgression studies; enables global ancestry analysis
Permutation Testing Frameworks [6]	Statistical significance assessment	Determining detection p-values by shuffling trait values across genotypes

Introgression represents a fundamental evolutionary process with far-reaching implications for adaptation, speciation, and conservation. The comparative analysis of detection methods reveals that tree-based and SNP-based approaches offer complementary strengths and limitations. Tree-based methods generally provide more conservative error control and enhanced detection of weaker associations by leveraging phylogenetic information, while SNP-based approaches offer greater flexibility for incorporating covariates and lower computational demands. The choice between methodological frameworks should be guided by research objectives, genomic context, and available computational resources. As genomic technologies continue to advance, the integration of these approaches with machine learning and functional validation holds promise for unraveling the full evolutionary significance of introgression across the tree of life.

The detection of introgression—the transfer of genetic material between species or populations through hybridization—is fundamental to understanding evolutionary history. Methods for identifying introgression broadly fall into two categories: tree-based approaches that use phylogenetic trees and SNP-based approaches that operate directly on genetic variants. Among SNP-based methods, the ABBA-BABA test and its corresponding Patterson's D statistic have become cornerstone techniques in evolutionary genomics. These methods quantify patterns of allele sharing to infer historical gene flow, providing a computationally efficient framework applicable to genome-scale data. This guide examines the SNP-based paradigm, detailing its methodologies, performance, and practical implementation in comparison with tree-based alternatives, providing researchers with the evidence needed to select appropriate methods for their specific research contexts.

Methodological Foundations of ABBA-BABA Tests

Core Principles and Statistical Framework

The ABBA-BABA test, formally known as Patterson's D statistic, operates on a quartet of populations or species: P1, P2, P3, and an outgroup (O). The test is built upon the principle that under a scenario of no gene flow, with the assumed phylogenetic relationship (((P1,P2),P3),O), two specific discordant allele patterns occur at equal frequencies due solely to incomplete lineage sorting (ILS). The "ABBA" pattern occurs when P2 and P3 share a derived allele while P1 retains the ancestral allele, and the "BABA" pattern occurs when P1 and P3 share the derived allele while P2 retains the ancestral allele [8] [9].

The D statistic quantifies the deviation from the expected equal occurrence of these patterns:

D = (Number of ABBA sites - Number of BABA sites) / (Number of ABBA sites + Number of BABA sites)

A D-statistic significantly different from zero indicates an excess of either ABBA or BABA sites, providing evidence of introgression. Specifically, a positive D value suggests gene flow between P2 and P3, while a negative D value suggests gene flow between P1 and P3 [8] [9]. Statistical significance is typically assessed using a Z-score based on block jackknifing, with |Z| > 3 often considered significant [9] [10].

Key Assumptions and Evolutionary Context

The ABBA-BABA test relies on several critical assumptions. First, it assumes the phylogenetic relationships among the four taxa are correctly specified. Second, it assumes identical substitution rates across lineages and the absence of homoplasies (recurrent mutations), meaning shared derived alleles result from common ancestry rather than independent mutations [11] [8]. These assumptions generally hold well for recently diverged species but may be problematic for more divergent taxa where rate variation and recurrent mutations become more likely [11].

Table 1: Key Assumptions of the ABBA-BABA Test

Assumption	Description	Potential Violations
Correct Topology	The relationship (((P1,P2),P3),O) must be correct	Incorrect phylogenetic placement of taxa
Clock-like Evolution	Equal substitution rates across lineages	Rate variation among species
No Homoplasy	No recurrent or back mutations	Multiple independent mutations at same site
Biallelic Sites	SNPs are biallelic	Multi-allelic sites, sequencing errors
Informative Sites	Only ABBA and BABA patterns inform the test	Other phylogenetic discordance patterns

Experimental Implementation and Protocols

Standard Workflow for D-Statistic Calculation

Implementing the ABBA-BABA test requires careful data preparation and analysis. The standard workflow begins with a Variant Call Format (VCF) file containing genomic polymorphisms and a population/species map specifying which individuals belong to which groups. The Dsuite software package provides an efficient implementation for genome-scale calculations directly from VCF files [8] [10].

The basic command structure in Dsuite for calculating D statistics across all population trios is:

Where the SETS.txt file is a tab-delimited text file with sample names and their corresponding population/species assignments, including the outgroup designation [10]. For studies involving many populations, the number of tests grows rapidly (as n choose 4), making computational efficiency an important consideration [8].

Advanced Analytical Extensions

Beyond the basic D statistic, several related statistics provide additional insights. The f4-ratio estimates the proportion of admixture in a population, while window-based statistics like fd, fdM, and df help identify specific introgressed loci by scanning along chromosomes [8]. The f-branch statistic (fb(C)) helps interpret systems of f4-ratio results across many populations by assigning evidence of gene flow to specific branches on a phylogeny, formalizing approaches used in studies of Heliconius butterflies [8].

For investigating significant signals of introgression in specific genomic regions, Dsuite provides the Dinvestigate command:

This command calculates fd, fdM, and df statistics in windows along the genome, allowing researchers to pinpoint regions potentially affected by introgression [10].

Performance Comparison: SNP-Based vs. Tree-Based Methods

Statistical Power and Error Control

Comparative studies reveal distinct performance characteristics between SNP-based and tree-based introgression detection methods. Tree-based methods generally demonstrate better control of Type I error rates (false positives) compared to non-tree-based methods. In one simulation study, a tree-based likelihood score statistic (LSS) showed error rates below 0.05, while a conventional t-test approach showed inflated error rates exceeding 0.05 across multiple genes [6].

For detection power, both approaches perform similarly well with strong genetic signals. However, in scenarios with weaker signals, tree-based methods that incorporate phylogenetic information may have advantages in localization—identifying SNPs closer to the true causal variants [6]. Tree-based methods also offer particular robustness to certain assumptions; they can provide reliable verification of introgression signals detected by SNP-based methods, especially when the assumption of equal substitution rates is violated [11].

Table 2: Performance Comparison of Introgression Detection Methods

Performance Metric	SNP-Based Methods (ABBA-BABA)	Tree-Based Methods
Type I Error Control	Can be inflated in some cases	Generally better controlled
Power with Strong Signals	High	High
Power with Weak Signals	Moderate	Potentially higher
Localization Accuracy	Moderate	Generally better
Computational Efficiency	High	Moderate to high
Handling Rate Variation	Problematic	More robust

Applicability Across Evolutionary Scenarios

The choice between SNP-based and tree-based methods depends heavily on the specific evolutionary context. ABBA-BABA tests are particularly well-suited for studies of recently diverged species or populations where the key assumptions of clock-like evolution and minimal homoplasy are reasonable [11] [8]. They also excel in applications requiring screening of many population combinations across whole genomes due to their computational efficiency [8] [10].

Tree-based methods demonstrate advantages in more complex evolutionary scenarios, including when analyzing divergent species with potential rate variation, when detailed phylogenetic information is available, and when seeking to corroborate signals detected by SNP-based methods [11] [6]. The robustness of tree-based methods to violations of the rate assumption makes them valuable for verifying introgression signals across diverse taxonomic groups [11].

Research Toolkit and Reagent Solutions

Essential Software Implementations

Table 3: Key Software Tools for Introgression Analysis

Tool	Primary Function	Key Features	Implementation
Dsuite [8] [10]	D-statistics and related analyses	Fast calculation from VCF files; implements D, f4-ratio, fd, fdM, f-branch	C++ with Python utilities
ADMIXTOOLS [8]	D-statistics and f4-ratio	Historically significant; comprehensive suite of statistics	C++ with Perl/R wrappers
Tree-based Pipeline [11]	Phylogenetic introgression detection	IQ-TREE for gene trees; ASTRAL for species tree; PhyloNet for networks	Multiple software integration
PhyloNet [11]	Species networks inference	Models reticulate evolution; implements maximum likelihood, Bayesian approaches	Java
ASTRAL [11]	Species tree from gene trees	Multi-species coalescent model; handles incomplete lineage sorting	Java

Data Requirements and Input Specifications

Successful implementation of ABBA-BABA tests requires properly formatted input data. The primary requirement is a VCF file containing biallelic SNPs, which may be compressed. While multiallelic loci and indels may be present in the VCF, only biallelic SNPs will be used in the analysis [10]. Additionally, a population/species map is required—a tab-delimited text file specifying which individuals belong to which populations, with the outgroup clearly designated using the "Outgroup" keyword [10].

For tree-based methods, a Newick-formatted tree may be required, with leaf labels matching population names in the dataset. Branch lengths may be included but are not always utilized depending on the specific analysis [11] [10]. When working with whole-genome alignments rather than VCFs, tools for extracting alignment blocks suitable for phylogenetic analysis are necessary, often requiring filtering based on information content and recombination signals [11].

Visualization of Method Workflows

ABBA-BABA Test Logic and Implementation

ABBA-BABA Test Workflow

Comparative Method Selection Framework

Method Selection Framework

The ABBA-BABA test and Patterson's D statistic represent powerful, efficient approaches for detecting introgression from genomic data, particularly well-suited for screening large datasets and studying recently diverged taxa. Their computational efficiency and straightforward implementation have made them indispensable tools in evolutionary genomics. However, their performance is contingent on specific evolutionary assumptions, particularly regarding rate uniformity and the absence of homoplasy.

For comprehensive introgression analysis, researchers should consider a hierarchical approach: beginning with efficient SNP-based methods like Dsuite to screen for signals across many population combinations, then applying more computationally intensive tree-based methods to verify significant findings, particularly when analyzing divergent taxa or when assumptions of rate constancy may be violated. This integrated methodology leverages the respective strengths of both paradigms, providing more reliable inferences about historical gene flow and its impact on evolutionary processes.

As genomic datasets continue expanding across diverse taxa, both SNP-based and tree-based methods are evolving, with recent developments including machine learning approaches that may offer additional insights in complex evolutionary scenarios [12]. Nevertheless, the ABBA-BABA test remains a foundational method that continues to provide critical insights into patterns of introgression across the tree of life.

In the field of genetic analysis, linking genomic variation to observable traits—a process fundamental to disease research and drug development—relies heavily on robust statistical methods. Two primary classes of association mapping methods have emerged: those that explicitly account for the evolutionary relatedness (phylogenetic tree-based methods) among individuals and those that ignore these evolutionary relationships (non-tree-based methods). Tree-based methods leverage the correlation structure imposed by shared evolutionary history, which can provide greater power to detect associations, particularly for traits with complex genetic architectures or weak effect sizes [6]. This guide provides an objective comparison of these approaches, focusing on their performance in detection power, type I error control, and localization accuracy, with supporting experimental data from genomic studies.

Performance Comparison: Tree-Based vs. Non-Tree-Based Methods

Direct comparisons of tree-based and non-tree-based methods reveal critical differences in their operational characteristics. The following table summarizes quantitative performance data from a controlled simulation study analyzing genes with varying effect sizes on systolic blood pressure [6].

Table 1: Performance comparison of a tree-based method (Likelihood Score Statistic - LSS) and a non-tree-based method (pooled t-test) across genes with different effect sizes.

Gene (Effect Size Magnitude)	Metric	Tree-Based Method (LSS)	Non-Tree-Based Method (t-test)
TNN (Large: up to 10.89)	Detection Power	High	High
LEPR (Large: up to 11.99)	Detection Power	High	High
FLT3 (Medium: up to 3.89)	Detection Power	Low (Similar performance)	Low (Similar performance)
TCIRG1 (Medium: up to 3.38)	Detection Power	Low (Similar performance)	Low (Similar performance)
GSN (Small: up to 0.76)	Detection Power	Low (Similar performance)	Low (Similar performance)
All Five Genes	Type I Error Rate (Target 0.05)	0.010 - 0.045 (Conservative)	> 0.05 (Slightly Inflated)

For detection power, both methods perform equally well in identifying genes with large effect sizes and show similarly low power for genes with small to medium effect sizes [6]. However, a key differentiator is type I error control—the probability of falsely detecting a non-existent association. The tree-based Likelihood Score Statistic (LSS) approach demonstrated conservative type I error rates (below the 0.05 target), whereas the classical t-test showed inflated error rates above 0.05 across all genes analyzed [6]. This suggests that tree-based methods may provide more reliable inference by reducing false positives.

Beyond genetic trait mapping, phylogenetic tree-based methods are also critical for resolving evolutionary relationships. A 2025 study comparing three phylogenetic methods using mitochondrial genomes of barnacles found that concatenated protein-coding genes (PCGs) significantly outperformed both gene order analysis and a single-marker (COX1) approach in preserving established taxonomic relationships (78.8% monophyly preservation vs. 50.0% and 61.3%, respectively) [13]. Furthermore, the trees generated by these different methods showed significant topological differences (Robinson-Foulds distances of 0.55–0.92), highlighting that methodological choice strongly influences phylogenetic conclusions [13].

Experimental Protocols and Workflows

Protocol for Tree-Based Association Mapping

The following workflow details the methodology for a tree-based association mapping study as described in [6]:

• Data Preparation and Quality Control:

  • Obtain genotypic data (e.g., from the Genetics Analysis Workshop 19 [6]).
  • Use a tool like Beagle to impute missing single nucleotide polymorphism (SNP) data and phase genotypic data into haplotypes.
  • Extract SNPs in genomic regions of interest (e.g., specific genes plus flanking regions) using software like VCFtools. Exclude SNPs lacking two or more variants across samples.

• Phylogenetic Tree Estimation:

  • At each SNP, estimate a phylogenetic tree (topology) from the haplotype data. This can be done using methods that leverage information from neighboring SNPs, such as the approach from Mailund et al. [6].
  • To reduce computational expense, use a broad-scale estimate of the tree. The tree is considered as a set of k clusters defined by the earliest (k-1) splits in the tree.

• Model Fitting and Statistical Testing:

  • Assume the trait values follow a multivariate normal distribution. The mean structure is defined by the cluster-specific mean trait values (μ₁, μ₂, ..., μₖ). The covariance structure, V(Θ), is defined by the estimated clustered tree, where the covariance between two observations is proportional to the length of their shared evolutionary branches [6].
  • Calculate the Likelihood Score Statistic (LSS) as a penalized likelihood measure: LSSᵢ = maxₖ [ 2 ln L(μ̂, σ̂² | y, V(Θ), Θ) - k ln n ] where μ̂ and σ̂² are maximum likelihood estimates, k is the number of clusters, and n is the number of observations (twice the number of individuals) [6].
  • The model is fitted for a range of cluster numbers (e.g., k=2 to k_max=15), and the score is based on the model with the maximum penalized likelihood.

• Significance Testing via Permutation:

  • To compute detection p-values, create permutation data sets (e.g., 100 sets) by shuffling trait values across genotypes, breaking any true genotype-phenotype associations.
  • The empirical p-value is the proportion of permuted data sets that produce a test statistic more extreme than the observed test statistic [6].

Protocol for Phylogenetic Tree Construction from Genomes

For phylogenetic inference from mitochondrial or nuclear genomes, the following protocol, derived from [13], applies:

• Sample Collection and DNA Sequencing:

  • Collect biological samples from the species of interest.
  • Extract genomic DNA and prepare a genomic library using a kit (e.g., QIAseq FX Single Cell DNA Library Kit).
  • Perform next-generation sequencing (e.g., on an Illumina NovaSeq 6000 system).

• Genome Assembly and Annotation:

  • Perform quality control on raw reads using tools like Trim_Galore to remove adapter sequences and low-quality data.
  • Assemble the mitochondrial genome using a combined de novo and reference-based approach with software like MitoZ.
  • Polish the assembly with a tool like Polypolish and annotate the genome to identify protein-coding genes (PCGs), rRNAs, and tRNAs.

• Dataset Compilation:

  • Compile a dataset of complete genomes from the study species and relevant outgroup species, often sourced from public repositories like NCBI GenBank.

• Multiple Sequence Alignment:

  • For nucleotide-based trees, align the sequences (e.g., concatenated PCGs or the COX1 marker) using an aligner like CLUSTAL Omega within Geneious Prime software.

• Phylogenetic Tree Construction:

  • For Gene-Order Analysis: Use a tool like Maximum Likelihood for Gene-Order (MLGO) to construct a tree based on the arrangement and orientation of all mitochondrial genes [13].
  • For Nucleotide Sequences: Use maximum likelihood software like raxmlGUI with the best-fitting nucleotide substitution model (e.g., GTR model). Node support should be assessed using a large number of bootstrap replicates (e.g., 1,000) [13].

The Scientist's Toolkit: Essential Research Reagents & Software

Successful implementation of tree-based genomic analyses requires a suite of specialized tools and reagents. The following table catalogues key solutions referenced in the experimental protocols.

Table 2: Essential research reagents and software for tree-based genomic analyses.

Category	Item Name	Function / Application
Wet-Lab Reagents	DNeasy Blood & Tissue DNA Kit (Qiagen)	Genomic DNA extraction from biological samples [13].
	QIAseq FX Single Cell DNA Library Kit (Qiagen)	Preparation of genomic libraries for next-generation sequencing [13].
	NovaSeq X Series Reagent Kit (Illumina)	Reagents for high-throughput sequencing on Illumina platforms [13].
Bioinformatics Software	Beagle	Imputation of missing SNP data and phasing of genotypic data into haplotypes [6].
	VCFtools	Processing and filtering of variant call format (VCF) files, e.g., extracting SNP data [6].
	MitoZ	De novo assembly and annotation of mitochondrial genomes [13].
	Trim Galore	Quality control and adapter trimming of raw sequencing reads [13].
	CLUSTAL Omega (within Geneious Prime)	Multiple sequence alignment of nucleotide or amino acid sequences [13].
	raxmlGUI / RAxML	Maximum likelihood phylogenetic tree construction from sequence alignments [13].
	Maximum Likelihood for Gene-Order (MLGO)	Phylogenetic tree construction based on gene order and rearrangement data [13].
Statistical Platforms	R (with phangorn, ape packages)	Statistical computing environment for phylogenetic comparison, calculating Robinson-Foulds distances, and monophyly tests [13].

The comparative data indicates that the choice between tree-based and non-tree-based methods is not a matter of one being universally superior. For detecting genetic associations with large effect sizes, simpler methods may suffice. However, tree-based approaches offer distinct advantages in controlling false positives (Type I error) and in explicitly modeling the evolutionary correlations inherent in population genetic data [6]. Furthermore, in phylogenetic studies, the choice of genomic data (e.g., concatenated PCGs vs. gene order) profoundly impacts the resulting evolutionary tree and its concordance with established taxonomy [13]. The decision must therefore be guided by the specific research question, the genetic architecture of the trait, the available genomic data, and the importance of robust error control in the inference process.

In the era of phylogenomics, reconstructing the evolutionary history of species has proven to be more complex than initially anticipated. Widespread gene tree discordance—the phenomenon where different genomic regions tell conflicting evolutionary stories—has emerged as a central challenge. This incongruence arises from various biological processes including incomplete lineage sorting (ILS), hybridization, and introgression, as well as methodological artifacts. Phylogenomic studies of diverse groups, from rattlesnakes and oaks to Asian columbines, consistently reveal that evolutionary histories are often not strictly tree-like but are better represented by networks that capture these complex relationships [14] [15] [16].

To address these challenges, researchers increasingly rely on sophisticated computational tools that can distinguish between different sources of phylogenetic conflict. This guide provides a comprehensive comparison of four essential software tools—ASTRAL, PhyloNet, IQ-TREE, and D-Suite—that form the core of modern phylogenomic analysis pipelines. We examine their performance characteristics, experimental applications, and appropriate use cases within the critical context of comparing tree-based versus SNP-based approaches for detecting introgression and other evolutionary processes.

The table below summarizes the key characteristics and primary applications of the four tools covered in this guide.

Table 1: Core Software Tools for Phylogenomic Analysis

Tool	Primary Function	Methodological Basis	Input Requirements	Key Outputs
ASTRAL	Species tree inference	Coalescent-based summary method	Collection of gene trees	Species tree with support values, branch lengths
PhyloNet	Reticulate evolution analysis	Multi-species coalescent networks	Gene trees or sequence alignments	Phylogenetic networks, introgression scenarios
IQ-TREE	Gene tree inference	Maximum likelihood phylogenetics	Sequence alignments	Gene trees, branch supports, model fit statistics
D-Suite	Introgression detection	D-statistic (ABBA-BABA) and related tests	Genotype data (VCF/PLINK)	D-statistics, f4-ratio tests, introgression graphs

Performance Comparison: Experimental Data and Applications

Empirical Performance Across Diverse Taxonomic Groups

Recent phylogenomic studies across diverse organisms provide critical insights into the performance characteristics of these tools under various biological scenarios:

• Plant Systems (Fagaceae): A 2025 study examining phylogenetic discordance in oaks and related species implemented a comprehensive pipeline using IQ-TREE for gene tree estimation, ASTRAL for species tree inference, and D-Suite analogues for introgression detection. The research quantified that gene tree estimation error accounted for 21.19% of observed variation, while biological processes of ILS (9.84%) and gene flow (7.76%) contributed significantly to discordance patterns. This study highlights the importance of using multiple complementary approaches to disentangle sources of conflict [15].

• Rattlesnakes (Crotalus and Sistrurus): Research published in 2024 demonstrated that the evolutionary history of rattlesnakes is dominated by rapid speciation and frequent hybridization. The authors utilized ASTRAL for coalescent-based species tree estimation and PhyloNet to infer phylogenetic networks, finding that both ILS and introgression contributed significantly to the extensive gene tree heterogeneity observed. Their results explained why previous studies using simpler concatenation approaches produced conflicting phylogenetic hypotheses [14].

• Asian Columbines (Aquilegia): A 2025 population genomic study of cryptic radiation in Aquilegia species from Southwest China employed D-Suite-related approaches to detect introgression signals. Researchers identified 39 out of 43 introgression events occurred post-lineage formation, with standing variation and introgression from non-sister lineages contributing to rapid genetic divergence without obvious morphological differentiation [16].

Detection Power and Limitations in Simulation Studies

Benchmarking studies using simulated datasets provide controlled assessments of tool performance:

Table 2: Performance Characteristics in Simulated and Empirical Studies

Tool	Strength	Limitation	Optimal Use Case
ASTRAL	Statistical consistency under ILS; scales to thousands of genes	Assumes no gene flow; may produce incorrect trees with strong introgression	Species tree inference in radiations with deep coalescence
PhyloNet	Explicitly models both ILS and introgression; infers complex networks	Computationally intensive for large numbers of taxa or reticulations	Detecting hybridization in moderately-sized clades
IQ-TREE	Model selection automation; accuracy for single-locus phylogenies	Does not account for ILS or introgression in single-gene trees	Gene tree estimation with appropriate substitution models
D-Suite	Efficient for genome-scale SNP data; robust to some rate variation	Assumes constant substitution rates; limited to quartet-based tests	Genome-wide scan for introgression using SNP data

Experimental Protocols and Workflows

Standard Phylogenomic Pipeline for Introgression Detection

The following diagram illustrates a comprehensive workflow integrating all four tools for phylogenomic analysis and introgression detection:

Detailed Methodological Protocols

Tree-Based Introgression Detection Protocol

Based on established workshops and recent publications, the tree-based detection protocol involves these critical steps [11]:

• Data Preparation and Alignment

  • Extract alignment blocks from whole-genome sequencing data
  • Filter blocks based on missing data and recombination breakpoints
  • Use tools like BWA for read mapping and GATK for variant calling

• Gene Tree Estimation

  • Employ IQ-TREE with command: iqtree2 -s alignment.phy -m MFP -B 1000
  • Use ModelFinder to select optimal substitution model
  • Assess node support with ultrafast bootstrap approximation

• Species Tree and Network Inference

  • Run ASTRAL with command: java -jar astral.5.7.8.jar -i genetrees.tre -o species.tre
  • Execute PhyloNet analyses for network inference: java -jar PhyloNet.jar script.net

• Concordance Analysis

  • Compare gene tree frequencies across the species tree
  • Identify asymmetries in quartet relationships suggestive of introgression

SNP-Based Introgression Detection Protocol

The SNP-based approach follows this general workflow [17] [16]:

• Variant Dataset Preparation

  • Generate genome-wide SNP datasets using standardized pipelines
  • Filter for missing data, quality scores, and minor allele frequency
  • Convert to appropriate formats (VCF, PLINK)

• Population Structure Assessment

  • Perform PCA using EIGENSOFT/SmartPCA
  • Run ADMIXTURE with varying K-values
  • Construct neighbor-joining trees for initial phylogenetic assessment

• Introgression Tests

  • Implement D-statistics (ABBA-BABA tests) using D-Suite
  • Calculate f4-ratio estimates to quantify introgression proportions
  • Perform phylogenetic scans for localized introgression signals

Tree-Based vs. SNP-Based Approaches: A Comparative Framework

The fundamental differences between tree-based and SNP-based approaches for introgression detection can be visualized as follows:

Key Methodological Distinctions

• Data Requirements: Tree-based methods utilize sequence alignments, preserving full phylogenetic information, while SNP-based approaches rely on genotype calls that represent genetic variation more compactly [11] [17].

• Underlying Assumptions: SNP-based D-statistics assume constant substitution rates and minimal homoplasy, which may be violated in divergent taxa. Tree-based approaches using sequence evolution models can accommodate rate variation and homoplasy through more complex models [11].

• Scalability and Resolution: D-Suite and related SNP-based tools efficiently handle genome-scale datasets but typically analyze four taxa at a time. Tree-based methods in PhyloNet can model complex networks but become computationally challenging with many taxa or extensive reticulation [11] [14].

Essential Research Reagent Solutions

The table below catalogues critical computational tools and resources that support phylogenomic analyses involving ASTRAL, PhyloNet, IQ-TREE, and D-Suite.

Table 3: Essential Computational Tools for Phylogenomic Analysis

Tool Category	Specific Software	Function in Workflow	Application Context
Sequence Alignment	BWA-MEM, Bowtie2	Read mapping to reference genomes	Pre-processing of WGS data for variant calling or alignment extraction [18]
Variant Calling	GATK, SAMtools	SNP and indel identification from aligned reads	Preparing genotype data for D-Suite analyses [16] [18]
Multiple Sequence Alignment	MAFFT, MUSCLE	Aligning homologous sequences	Creating input alignments for IQ-TREE [11]
Population Genetics	PLINK, ADMIXTURE	Population structure analysis	Complementary analysis for interpreting introgression signals [18]
Tree Visualization	FigTree, IcyTree	Visualization and annotation of phylogenetic trees	Exploring and presenting results from ASTRAL, IQ-TREE [11]
Simulation Tools	ALF, Dawg	Simulating genome evolution under complex models	Benchmarking tool performance under known evolutionary scenarios [19]

Based on comparative analyses across numerous empirical studies, the most effective strategy for comprehensive introgression detection involves integrating both tree-based and SNP-based approaches. Tree-based methods using PhyloNet and ASTRAL provide powerful frameworks for modeling complex evolutionary histories that incorporate both ILS and introgression, while SNP-based tools like D-Suite offer efficient genome-wide scans for introgression signals. IQ-TREE serves as a critical component for accurate gene tree estimation underlying both approaches.

Future methodology development will likely focus on improving scalability of network approaches, better integration of comparative genomics and population genetic approaches, and developing more robust statistical frameworks that jointly model multiple sources of phylogenetic conflict. As demonstrated across diverse biological systems, from oaks and pines to rattlesnakes and cattle, combining these complementary approaches provides the most comprehensive understanding of evolutionary history and the role of introgression in adaptation and diversification.

The accurate detection of introgression, the transfer of genetic material between species or populations through hybridization and repeated backcrossing, is fundamental to understanding evolution, local adaptation, and speciation. As genomic data becomes increasingly abundant, two primary computational approaches have emerged for identifying introgressed sequences: tree-based methods and SNP-based methods. Each paradigm offers distinct advantages and faces specific limitations. This guide provides an objective comparison of their performance, supported by experimental data and detailed protocols, to assist researchers in selecting the appropriate tool for their specific research context in evolutionary biology and drug development.

The table below summarizes the core performance characteristics of tree-based and SNP-based introgression detection methods, synthesizing findings from current research.

Table 1: Comparative Performance of Introgression Detection Methods

Metric	Tree-Based Methods	SNP-Based Methods (e.g., D-statistic/ABBA-BABA)
Fundamental Principle	Compares gene tree topologies and frequencies across the genome to a known species tree [11].	Compares patterns of derived allele sharing (e.g., ABBA vs. BABA sites) to detect asymmetry indicative of gene flow [11].
Key Assumptions	Fewer assumptions about evolutionary rates; models sequence evolution explicitly [11].	Assumes identical substitution rates and absence of homoplasy (multiple independent substitutions) [11].
Optimal Use Case	Divergent species complexes and scenarios involving complex demographic histories [11].	Recently diverged species groups with low rates of homoplasy [11].
Robustness to Homoplasy	High, as phylogenetic methods model or are less misled by multiple hits [11].	Low, as homoplasy can produce false-positive signals of introgression [11].
Computational Demand	High (requires building many gene trees) [11].	Low (fast calculation on SNP data).
Output	Set of gene trees; visualization of introgression in a phylogenetic network [11].	A single statistic (e.g., D) quantifying the deviation from a strict bifurcating tree [11].

Experimental Protocols

To ensure reproducibility and provide a clear framework for performance testing, this section outlines detailed protocols for implementing both tree-based and SNP-based analyses, as applied in recent studies.

Tree-Based Introgression Workflow

The following protocol, adapted from a population genomics workshop, details the steps for a tree-based analysis using a whole-genome alignment [11].

• Data Preparation: Whole-Genome Alignment

  • Input: Genome assemblies for the target species and an outgroup.
  • Method: Generate a chromosome-scale whole-genome alignment using a tool like Progressive Cactus. The example dataset consists of five cichlid species (Neolamprologus spp.) and an outgroup (Nile tilapia), aligned to a single chromosome [11].
  • Format: The alignment can be converted to the human-readable MAF (Multiple Alignment Format) for inspection using tools like hal2maf [11].

• Extraction and Filtering of Alignment Blocks

  • Objective: Isolate blocks suitable for phylogenetic inference by minimizing missing data and within-alignment recombination.
  • Procedure:
    • Use a custom Python script to extract alignment blocks of a fixed length (e.g., 1,000 bp) from the whole-genome alignment.
    • Filter blocks to retain only those containing one sequence for every species in the analysis.
    • Quantify signals of recombination per alignment and remove blocks with the strongest signals [11].

• Gene Tree Inference

  • Tool: IQ-TREE (v.2.0+)
  • Method: Perform maximum likelihood phylogenetic inference on each filtered alignment block. Use model selection to find the best-fit nucleotide substitution model for each block [11].
  • Output: A set of thousands of gene trees, each in Newick format.

• Species Tree and Introgression Analysis

  • Species Tree Estimation: Use ASTRAL to estimate a consensus species tree from the entire set of gene trees. This method is efficient and accounts for incomplete lineage sorting [11].
  • Introgression Detection:
    • Topology Frequency Analysis: Examine the distribution of gene tree topologies. An excess of trees that do not match the species tree topology can indicate introgression.
    • Phylogenetic Network Inference: Use PhyloNet to infer a species network that directly models introgression events as horizontal edges between lineages [11].

SNP-Based Introgression Workflow

This protocol summarizes the SNP-based approach, highlighting its application in a study on East and Southeast Asian populations [20].

• Data Preparation: Genotype Calling and Quality Control

  • Input: Whole-genome or reduced-representation sequencing data (e.g., from SNP arrays).
  • Quality Control (QC): Use PLINK 1.9 to filter samples and SNPs.
    • Remove samples with >10% missing genotypes (--mind 0.1).
    • Remove SNPs with >10% missing call rate (--geno 0.1).
    • Exclude SNPs with a minor allele frequency (MAF) < 1% (--maf 0.01).
    • Apply a Hardy-Weinberg equilibrium (HWE) filter (p-value < 0.001, --hwe 0.001) [20].
  • Output: A high-quality set of biallelic SNPs (e.g., 597,569 SNPs retained in the Asian population study) [20].

• Population Structure Analysis

  • Tool: ADMIXTURE
  • Method: Perform unsupervised clustering to estimate individual ancestry proportions across a range of K (number of ancestral populations). Use cross-validation (e.g., --cv=10) to identify the most supported K value [20].

• Selection of Ancestry-Informative SNPs (AISNPs)

  • Objective: Identify a reduced panel of SNPs with high power to distinguish populations.
  • Tool: AIM generator or similar.
  • Method: Rank SNPs using a statistic like Rosenberg's In, which measures allelic frequency differentiation between populations. Select a nested panel (e.g., 50 to 2,000 SNPs) based on chromosomal distribution and linkage disequilibrium (LD) pruning [20].

• Introgression Detection with D-statistics

  • Principle: The ABBA-BABA test compares patterns of shared derived alleles between four populations (((P1, P2), P3), Outgroup).
  • Calculation: A significant excess of ABBA or BABA site patterns indicates gene flow between P3 and P2 or P3 and P1, respectively. The D-statistic quantifies this deviation [11].

• Ancestry Classification (Alternative/Complementary Approach)

  • Tool: Machine learning classifiers (e.g., XGBoost, Random Forests, CNN).
  • Method: Train a model on reference genotype data (e.g., the AISNP panel) with known population labels. The optimized XGBoost model achieved 95.6% accuracy with 2,000 AISNPs in one study [20].
  • Application: Classify unknown or admixed individuals to populations, which can infer historical introgression.

Case Studies and Experimental Data

Case Study 1: Tree-Based Detection in Cichlid Fishes

A study on five Neolamprologus cichlid species used a tree-based approach to verify signals of introgression.

• Method: Researchers extracted 1,000 bp alignment blocks from a whole-genome alignment and inferred over 10,000 gene trees using IQ-TREE.
• Finding: The distribution of gene-tree topologies showed significant asymmetry, supporting past introgression events between specific species pairs. This analysis served to verify results from SNP-based tests under conditions where homoplasy could have misled the latter [11].

Case Study 2: SNP-Based Detection in Pines

A large-scale genomic study on Pinus sylvestris and P. mugo utilized SNP data to investigate adaptive introgression.

• Experimental Data:
  • Samples: 1,558 trees from 24 allopatric and hybrid-zone populations.
  • Genotyping: Thousands of nuclear SNPs were genotyped.
  • Analysis: Population structure and ancestry assignment revealed groups of pure species, F1 hybrids, and advanced backcrosses. A majority of hybrids showed a genetic shift towards P. mugo ancestry, indicating asymmetric introgression [17].
• Finding: Outlier tests identified SNPs under selection, many shared across different hybrid zones. These SNPs were linked to regulatory processes like phosphorylation and transmembrane transport, suggesting that adaptive introgression is facilitating the exchange of beneficial alleles for stress tolerance [17].

Case Study 3: Adaptive Introgression in Poplar Trees

A 31-year common garden experiment with Populus fremontii and P. angustifolia provided direct evidence for climate change resilience via introgression.

• Experimental Data:
  • Design: Genotypes of both parental species, F1 hybrids, and backcrosses were planted in a warm, low-elevation common garden.
  • Survival: After 31 years, ~90% of P. fremontii and 100% of F1 hybrids survived, compared to only ~25-30% of P. angustifolia and backcrosses.
  • Selection Pressure: For the vulnerable P. angustifolia and backcrosses, each 1°C increase in transfer distance (source vs. garden temperature) decreased odds of survival by 7.5% [21].
• Finding: Among the surviving P. angustifolia and backcross trees, the presence of specific introgressed genetic markers (e.g., RFLP-1286 from P. fremontii) was associated with a 75% greater survival rate, demonstrating a direct marker-trait association for climate resilience [21].

Visualized Workflows

The following diagrams illustrate the logical workflows for tree-based and SNP-based introgression detection, providing a clear overview of the analytical pipelines.

Tree-Based Introgression Analysis Workflow

SNP-Based Introgression Analysis Workflow

The Scientist's Toolkit

This section details essential research reagents, software, and data sources critical for conducting introgression analyses.

Table 2: Essential Research Reagents and Solutions for Introgression Studies

Category	Item/Tool	Function and Application
Bioinformatics Software	IQ-TREE	Efficient maximum likelihood inference of phylogenetic trees from molecular sequences; used for generating gene trees [11].
	ASTRAL	Estimates the primary species tree from a set of input gene trees, accounting for incomplete lineage sorting [11].
	PhyloNet	Infers phylogenetic networks to model and visualize evolutionary relationships that include reticulations like introgression and hybridization [11].
	PLINK	A whole-genome association analysis toolset used for rigorous quality control, manipulation, and filtering of SNP datasets [20].
	ADMIXTURE	A tool for estimating ancestry proportions and inferring population structure from genotype data in a maximum-likelihood framework [20].
Data Sources	Whole-Genome Alignment (HAL/MAF)	A reference-free or reference-based multiple genome alignment, serving as the input for tree-based methods to extract homologous blocks [11].
	VCF File	The Variant Call Format file storing sequence variations (SNPs, indels) for all samples, forming the primary input for SNP-based methods [20] [17].
Analytical Resources	Ancestry-Informative SNP (AISNP) Panels	A reduced set of SNPs with high power to differentiate populations; enables cost-effective and efficient ancestry analysis [20].
	Common Garden Experiments	Long-term experiments where genotypes from different environments are grown together; used to measure fitness and identify adaptive traits under controlled conditions [21].

A Practical Guide to Implementing Introgression Detection Tests

The precise identification of introgressed genomic regions—segments of DNA transferred between species or populations through hybridization and backcrossing—is fundamental to understanding evolutionary processes, local adaptation, and the genetic basis of complex traits. Single Nucleotide Polymorphisms (SNPs) serve as pivotal molecular markers in these investigations due to their abundance across genomes and role as signatures of historical evolutionary events [22] [23]. The workflow from whole-genome alignment to D-statistic calculation represents a cornerstone methodology for detecting introgression, providing a computational framework to distinguish true gene flow from other evolutionary forces such as incomplete lineage sorting [11] [24].

This guide objectively compares the performance of this established SNP-based workflow against emerging methodologies, particularly tree-based phylogenetic approaches. The comparative analysis is situated within the broader thesis that while SNP-based methods, especially those leveraging the ABBA-BABA D-statistic, offer powerful and accessible tests for introgression, they operate under specific assumptions that can be complemented by the phylogenetic signal captured by tree-based methods [11] [12]. The D-statistic quantifies the excess of shared derived alleles between populations, which is a key signal of introgression, but its accuracy depends on critical assumptions, including identical substitution rates and the absence of homoplasies (multiple independent substitutions at the same site), which are more likely to hold in recently diverged species [11].

Comparative Workflow Analysis: SNP-Based vs. Tree-Based Detection

The detection of introgression relies on distinguishing patterns of shared genetic variation resulting from gene flow from those caused by other evolutionary processes. The following workflows represent two dominant paradigms in the field.

The SNP-Based ABBA-BABA Workflow

The D-statistic, or ABBA-BABA test, is a widely used summary statistic-based method for detecting introgression. It tests for an imbalance in the patterns of shared derived alleles between four taxa (P1, P2, P3, and an outgroup) [11] [23]. The core workflow is outlined in the diagram below.

Experimental Protocol for D-Statistic Calculation:

• Sequence Alignment: Map sequencing reads from all populations under study to a high-quality reference genome using aligners like Bowtie2 or BWA [22] [24]. The reference should share high homology with the presumed ancestral lineage for optimal results.
• Variant Calling: Identify SNPs across all samples. This can be done using standard callers like the Genome Analysis Toolkit (GATK) or more specialized, scalable tools like Heap integrated with Hadoop for large datasets [22].
• Data Filtering: Apply quality filters (e.g., coverage depth, genotype quality, missing data) to obtain a high-confidence set of SNPs. Filtering is critical for reducing false positives in downstream analysis [25].
• Site Pattern Identification: For each SNP, polarize the alleles using an outgroup sequence to determine the ancestral (A) and derived (B) states. Classify sites into patterns:
  • ABBA Sites: P1 has the ancestral allele, P2 and P3 share the same derived allele.
  • BABA Sites: P2 has the ancestral allele, P1 and P3 share the same derived allele.
• D-Statistic Calculation: Compute the D-statistic using the formula:
  • ( D = (N{ABBA} - N{BABA}) / (N{ABBA} + N{BABA}) )
  • Where ( N{ABBA} ) and ( N{BABA} ) are the counts of ABBA and BABA sites, respectively. A significant deviation of D from zero indicates introgression between P2 and P3 (if D>0) or P1 and P3 (if D<0) [11] [23].
• Significance Testing: Assess the statistical significance of the D-value using a block jackknife or permutation test to account for the non-independence of linked SNPs.

The Tree-Based Introgression Detection Workflow

As a complementary approach, tree-based methods detect introgression by analyzing the distribution of gene tree topologies inferred from sequence alignments across the genome [11]. The workflow is illustrated below.

Experimental Protocol for Tree-Based Detection:

• Extract Alignment Blocks: From a whole-genome alignment, extract numerous, non-overlapping sequence blocks (e.g., 1,000 bp in length). These blocks are filtered for completeness, information content, and a low frequency of recombination breakpoints to ensure phylogenetic reliability [11].
• Infer Gene Trees: For each filtered alignment block, reconstruct a phylogenetic tree (a "gene tree") using maximum likelihood methods implemented in tools like IQ-TREE [11].
• Estimate Species Tree: Reconcile the individual gene trees to infer the dominant species tree topology using a coalescent-based method like ASTRAL, which is robust to incomplete lineage sorting [11].
• Analyze Topology Frequencies: Compare the distribution of the gene tree topologies to the expected distribution under the multispecies coalescent model without introgression. A significant excess of gene trees that match a specific alternative topology (e.g., one that groups P2 and P3) provides evidence for introgression between those taxa [11].
• Infer Species Networks (Optional): Use tools like PhyloNet to explicitly model and test different hybridization or introgression scenarios within a phylogenetic network framework [11].

Performance Comparison: Quantitative Data and Experimental Findings

Direct comparisons of these methodologies reveal distinct performance characteristics, strengths, and limitations. The table below summarizes key findings from experimental studies and benchmarks.

Table 1: Comparative Performance of Introgression Detection Methods

Methodological Feature	SNP-Based (D-Statistic)	Tree-Based (Gene Tree Topologies)	Supporting Evidence
Core Principle	Allele frequency imbalance (ABBA-BABA)	Distribution of gene tree topologies	[11] [23]
Key Assumptions	Identical substitution rates; no homoplasy	Model of sequence evolution; effective recombination between loci	[11]
Computational Intensity	Moderate (scales with number of SNPs)	High (scales with number of loci x complexity of tree inference)	[22] [11]
Handling of Deep Divergence	Problematic (violates assumptions)	Robust (explicitly models ancestral variation)	[11]
Output Granularity	Genome-wide or window-based test	Can localize introgression to specific genomic regions	[11] [23]
Accuracy in Simulation Studies	High for recent introgression with low homoplasy	High across diverse evolutionary scenarios, including deep divergence	[11]
Reported Application Example	Detecting TT1 and GLW7 gene introgression from indica to tropical japonica rice [23]	Analyzing introgression in Neolamprologus cichlid fishes of Lake Tanganyika [11]	[11] [23]

Performance data indicates that the D-statistic can be unreliable when its underlying assumptions are violated, such as in comparisons of highly divergent species where homoplasy is more likely [11]. In contrast, phylogenetic approaches that use full sequence alignment information can be more robust under these conditions, serving as a vital verification for SNP-based findings [11]. Furthermore, novel bioinformatics pipelines like IntroMap demonstrate alternative SNP-based approaches that avoid variant calling altogether, instead using signal processing of alignment data to detect introgressed regions with reported high accuracy in plant breeding applications [24].

The Scientist's Toolkit: Essential Research Reagents and Solutions

Successful implementation of introgression detection workflows relies on a suite of specialized software tools and computational resources.

Table 2: Essential Research Reagents and Computational Tools

Tool Name	Primary Function	Role in Workflow	Key Feature
Bowtie2 / BWA	Short-read alignment	Aligns NGS reads to a reference genome	Fast, memory-efficient mapping for whole-genome data [22] [24]
GATK / Heap	Variant discovery	Calls SNPs from aligned reads	GATK is industry-standard; Heap is Hadoop-based for scalability [22]
IQ-TREE	Phylogenetic inference	Infers maximum likelihood gene trees from alignment blocks	Modern, fast, and model-rich tree building [11]
ASTRAL	Species tree estimation	Estimates the primary species tree from a set of gene trees	Coalescent-based, accounts for incomplete lineage sorting [11]
PhyloNet	Phylogenetic network inference	Models and tests explicit introgression/hybridization scenarios	Infers evolutionary histories that are not strictly tree-like [11]
IntroMap	Introgression detection	Identifies introgressed regions from BAM alignments without variant calling	Signal processing-based; avoids potential biases from SNP calling [24]
PAUP*	Phylogenetic analysis	General utility for tree inference and manipulation (command-line used)	Legacy tool with comprehensive feature set for phylogenetic analyses [11]

The comparative analysis of SNP-based and tree-based introgression detection methods reveals a landscape defined by a trade-off between accessibility and robustness. The SNP-based D-statistic workflow provides a fast, scalable, and statistically powerful framework that is ideal for screening for gene flow in large genomic datasets, particularly among recently diverged populations [22] [23]. However, its performance is contingent upon evolutionary assumptions that are often violated in practice.

Emerging research underscores that tree-based methods offer a critical complementary approach. By leveraging the full information in sequence alignments and explicitly modeling phylogenetic histories, they are more robust to conditions like deep divergence and homoplasy that can confound the D-statistic [11]. The most rigorous studies in the field now often employ both approaches in tandem: using the D-statistic for initial genome-wide screening and tree-based methods to confirm specific introgression events and model complex evolutionary histories [11] [23]. This integrated methodology provides a more reliable and nuanced understanding of the genomic landscapes of introgression.

In the field of evolutionary biology, accurately reconstructing gene trees is fundamental for understanding the relationships between species, genes, and their evolutionary history. This process involves multiple critical steps: extracting homologous sequences, performing multiple sequence alignment, and conducting phylogenetic inference. With the availability of various phylogenetic tools, selecting the most appropriate software is crucial for obtaining reliable results. IQ-TREE has emerged as a widely used software for maximum likelihood phylogenomic inference, known for its speed, accuracy, and extensive model selection capabilities. This guide provides a comprehensive comparison of IQ-TREE's performance against alternative phylogenetic tools, with a specific focus on its application within broader research comparing tree-based methods with SNP-based introgression tests. We present experimental data and benchmarking studies to objectively evaluate these tools, providing researchers with evidence-based recommendations for their genomic analyses.

Experimental Protocols and Benchmarking Methodologies

To objectively evaluate the performance of phylogenetic tools, researchers typically employ standardized benchmarking protocols. These involve simulating sequence data under controlled evolutionary conditions and then comparing the accuracy of different inference methods in recovering the known "true" phylogeny.

Standard Phylogenetic Benchmarking Protocol:

• Data Simulation: Generate synthetic DNA or protein sequence alignments using evolutionary simulators that incorporate realistic mutation models, rate heterogeneity, and specified tree topologies. For B-cell receptor (BCR) sequences, specialized simulators model context-dependent somatic hypermutation and affinity-based selection [26].
• Tree Inference: Apply multiple phylogenetic inference tools (e.g., IQ-TREE, RAxML, PhyML, B-cell specific tools) to the simulated alignments.
• Accuracy Assessment: Compare the inferred trees to the simulated "true" tree using metrics like Robinson-Foulds distance (for topological accuracy) and F-score (for branch support). For ancestral sequence reconstruction, accuracy is measured by the percentage of correctly inferred ancestral nucleotides or amino acids [26].

SNP-Based Introgression Test Protocol: In contrast to full-sequence tree-building, SNP-based methods often rely on genotyping assays. A typical protocol for estimating introgression levels, as used in honeybee conservation, involves [5]:

• SNP Panel Design: Identify ancestry-informative markers from whole-genome data.
• High-Throughput Genotyping: Use technologies like the iPLEX MassARRAY system to genotype these SNPs across numerous samples.
• Admixture Analysis: Calculate introgression proportions (Q-values) using model-based clustering algorithms and compare the results against those derived from whole-genome data to assess the panel's accuracy.

Performance Comparison of Phylogenetic Tools

Benchmarking studies reveal significant variation in the performance of different phylogenetic methods, particularly when applied to specific data types like B-cell receptor sequences.

Table 1: Benchmarking Performance of Phylogenetic Tools on Simulated B-Cell Receptor Sequences [26]

Tool Category	Specific Tool	Key Features/Methodology	Inference Accuracy (Relative Performance)	Ancestral Sequence Reconstruction Accuracy
Classical Maximum Likelihood	RAxML, PhyML, IQ-TREE	General-purpose substitution models	Variable; can be suboptimal for BCR data	Lower than BCR-specific tools
Classical Maximum Parsimony	PHYLIP dnapars	Minimal number of evolutionary changes	Can be effective with limited divergence	Not specialized for BCR motifs
BCR-Specific Tools	IgPhyML	Adapts codon model for SHM motifs	High	High
	GCtree	Ranks parsimony trees with a branching process model	High (with single-cell data)	N/A
	SAMM	Ranks trees based on SHM motif likelihood	High	High

The data indicates that tools specifically designed to model the unique characteristics of B-cell receptor evolution, such as IgPhyML and SAMM, consistently outperform general-purpose phylogenetic software. This performance gain is attributed to their ability to account for context-dependent somatic hypermutation, a feature not captured by standard substitution models used in RAxML, PhyML, or IQ-TREE [26]. This highlights the importance of selecting a tool whose underlying model matches the biological process under study.

IQ-TREE: Features and Advanced Workflows

IQ-TREE is a versatile software that addresses several key challenges in phylogenomics. Its core strengths include sophisticated model selection and the ability to handle complex, partitioned data sets.

Key Features and Capabilities

• ModelFinder: This built-in function implements a fast model selection algorithm that chooses the best-fit substitution model from a large set of candidates to avoid overparameterization and improve inference accuracy [27].
• Partitioned Analysis: For multi-gene alignments, IQ-TREE allows users to define data partitions (e.g., different genes or codon positions) and assign separate substitution models to each. The -spp option is recommended, as it allows each partition to have its own evolution rate, providing a balance between realism and model complexity [27].
• Ultrafast Bootstrap Approximation: IQ-TREE offers an ultrafast bootstrap (UFBoot) algorithm that is significantly faster than the standard bootstrap while providing more unbiased support values [27].
• Tree Search Efficiency: IQ-TREE employs efficient heuristics for a thorough search of tree space, helping to find trees with high likelihood scores.

Workflow for Partitioned Analysis with IQ-TREE

A robust workflow for building a gene tree from a multi-locus alignment using IQ-TREE's partition model is outlined below.

This workflow allows IQ-TREE to automatically find the optimal partitioning scheme and model for a concatenated alignment, then infer a robust phylogenetic tree with branch support values.

Tree-Based vs. SNP-Based Methods for Introgression Analysis

The choice between using full sequence data for tree-building versus a reduced set of SNPs for introgression testing depends on the research goals, budget, and computational resources.

Table 2: Comparison of Tree-Based Phylogenetic and SNP-Based Introgression Methods

Aspect	Tree-Based Phylogenetic Methods (e.g., IQ-TREE)	SNP-Based Introgression Tests
Data Basis	Uses full sequence alignments (DNA, protein, or codons) [28].	Uses a panel of pre-selected, informative SNPs [5].
Primary Output	Phylogenetic tree showing evolutionary relationships and divergence.	Admixture proportions (Q-values) quantifying ancestry from different populations [5].
Key Strengths	Provides rich evolutionary context (orthologs/paralogs, ancestral states) [28]. High accuracy for inferring evolutionary relationships [28].	Fast and cost-effective for genotyping many samples [5]. Requires less computational power and bioinformatics expertise [5].
Limitations	Computationally intensive for large datasets. Requires careful model selection and alignment.	Limited to pre-defined questions (e.g., ancestry proportions). Reduced resolution compared to full sequence [5].
Typical Use Case	Deep evolutionary studies, gene family classification, ancestral sequence reconstruction [26] [28].	Population monitoring, conservation genetics, breeding programs [5].

The performance of SNP-based methods is highly dependent on the number and informativeness of the selected markers. For example, a study on honeybee conservation found that a panel of 117 SNPs could estimate introgression with an accuracy of 97.84% compared to whole-genome data, while a smaller panel of 62 SNPs still achieved over 96.9% accuracy, offering a good compromise between cost and precision [5]. This demonstrates that while full-sequence tree inference is more powerful for detailed evolutionary analysis, targeted SNP panels can be a highly efficient and accurate alternative for specific applications like introgression testing.

Table 3: Key Software and Analytical Tools for Phylogenetics and Introgression Analysis

Tool Name	Type	Primary Function	Relevance to Gene Tree/Introgression Research
IQ-TREE	Software Package	Maximum Likelihood Phylogenetic Inference	Core tool for building gene trees from sequence alignments with model selection and branch support [27].
SHOOT	Online Tool / Database	Phylogenetic Gene Search and Ortholog Inference	Rapidly places a query gene into a pre-computed gene tree, providing evolutionary context and orthologs [28].
IgPhyML	Software Package	BCR-Specific Phylogenetic Inference	Specialized for accurate tree and ancestral sequence inference from B-cell receptor data [26].
iPLEX MassARRAY	Genotyping Platform	High-Throughput SNP Genotyping	Enables cost-effective genotyping of customized SNP panels for introgression analysis in large sample sets [5].
DNALONGBENCH	Benchmark Dataset	Evaluation of Long-Range DNA Prediction	Standardized resource for assessing model performance on tasks requiring long sequence contexts [29].

Building robust gene trees requires careful consideration of each step in the phylogenetic pipeline, from data extraction to final inference. IQ-TREE stands out as a powerful and flexible tool for maximum likelihood analysis, particularly due to its sophisticated model selection, partition modeling capabilities, and efficient tree search algorithms. However, benchmarking evidence clearly shows that for specialized data types like B-cell receptor sequences, BCR-specific tools like IgPhyML can achieve superior accuracy by incorporating domain-specific evolutionary models [26].

The choice between a full tree-based method and a targeted SNP-based approach should be guided by the research question. For deep evolutionary analysis and gene family classification, tree-based methods with IQ-TREE are superior, providing a comprehensive phylogenetic context. For applied conservation genetics or breeding programs where cost and throughput are primary concerns, SNP-based introgression tests offer a highly efficient and accurate alternative [5]. Ultimately, leveraging benchmarked tools and validated experimental protocols, as detailed in this guide, will empower researchers to generate more reliable and biologically meaningful phylogenetic conclusions.

In the field of phylogenomics, accurately reconstructing species evolutionary history is complicated by processes that cause individual gene histories to differ from the species tree. Incomplete lineage sorting (ILS) and hybridization are two major sources of this gene tree discordance [30]. This guide focuses on two principal software approaches: ASTRAL, a leading method for species tree inference from gene trees under ILS, and PhyloNet, a comprehensive tool for inferring phylogenetic networks that explicitly represent hybridization and other reticulate evolutionary events. Understanding their comparative performance with alternative methods and SNP-based approaches is crucial for researchers investigating introgression and evolutionary relationships.

The Multi-Species Coalescent (MSC) and Beyond

The Multi-Species Coalescent (MSC) model provides a statistical framework for understanding gene tree discordance due to ILS, which occurs when ancestral genetic lineages fail to coalesce within a population divergence time [31]. The MSC models the probability distribution of gene trees within a species tree and serves as the foundational model for a class of methods known as "summary methods," which estimate species trees from a collection of input gene trees [31]. Methods like ASTRAL are statistically consistent under the MSC, meaning they converge to the true species tree given sufficient gene tree data [32] [31].

However, the MSC does not account for gene flow. When hybridization or introgression occurs, the evolutionary history is better represented by a phylogenetic network, which incorporates nodes of hybrid origin with multiple parents. PhyloNet is a software package specifically designed for inferring and analyzing such networks [30].

ASTRAL (Accurate Species TRee ALgorithm) estimates species trees by finding the tree that shares the largest number of induced quartet trees with the set of input gene trees [32]. Its statistical consistency under the MSC, scalability, and accuracy have made it one of the most widely used species tree methods. The recently introduced ASTER package consolidates ASTRAL and its variants (e.g., wASTRAL for weighted gene trees, ASTRAL-Pro for multi-copy genes, and CASTER for direct alignment input) into a unified toolkit [32].

PhyloNet is a package for analyzing phylogenetic networks. It provides commands for inferring networks from gene trees or sequence alignments (e.g., using maximum likelihood or parsimony), simulating gene tree evolution within networks, and comparing networks. It is particularly powerful for detecting hybridization and introgression events.

Table: Comparison of Core Methodologies

Feature	ASTRAL	PhyloNet
Primary Goal	Species tree inference	Phylogenetic network inference
Underlying Model	Multi-species Coalescent (ILS)	Multi-species Coalescent + Reticulation (ILS + Hybridization)
Standard Input	Gene tree topologies (Newick files)	Gene trees or multiple sequence alignments
Key Assumption	Discordance is primarily due to ILS	Discordance can be due to ILS and hybridization
Statistical Consistency	Consistent under MSC	Consistent under the network MSC model for certain algorithms

Comparative Landscape of Species Tree and Network Methods

Several other methods address species tree inference and hybridization detection. STELAR is a triplet-based species tree method that, like ASTRAL, is statistically consistent under the MSC and has been shown to match ASTRAL's accuracy [31]. SNP-based methods for introgression tests, such as D-statistics (ABBA-BABA tests) and f4-statistics, use patterns of allele sharing across genomes to detect signals of gene flow between taxa without explicitly inferring a network. A key comparative thesis is that tree-based methods like ASTRAL and PhyloNet model the underlying population genetic processes (coalescence) explicitly, while many SNP-based tests are primarily descriptive and identify patterns consistent with introgression.

Performance and Accuracy Comparison

Benchmarking ASTRAL and Its Alternatives

Experimental studies consistently demonstrate the high accuracy and scalability of coalescent-based summary methods.

• ASTRAL vs. STELAR and MP-EST: A comparative study showed that STELAR matches the accuracy of ASTRAL and improves upon MP-EST and SuperTriplets across a range of simulated conditions [31]. STELAR achieves this by solving the Constrained Triplet Consensus (CTC) problem, finding a species tree that agrees with the largest number of rooted triplets induced by the gene trees [31].
• Scalability of ASTER: The latest implementation, ASTRAL-IV within the ASTER package, shows remarkable scalability. It can infer a phylogeny of 363 bird species from 63,430 gene trees in just 2 hours using 32 CPU cores, a significant speedup over previous versions [32]. Furthermore, its improved handling of missing data makes it effective for super-tree estimation [32].
• CASTER for Direct Alignment Input: CASTER is a site-based method within ASTER that bypasses gene tree estimation entirely, inferring the species tree directly from a multiple sequence alignment. One benchmark showed CASTER is 800 times less CPU-intensive than a two-step method (maximum likelihood gene trees + weighted ASTRAL) for a 201-species, 10,000-loci dataset, while achieving higher accuracy under high ILS [32].

Table: Summary of Experimental Performance Data

Method / Tool	Key Performance Finding	Experimental Context
ASTRAL-IV	2 hours runtime (32 cores) for 363 species and 63,430 genes [32]	Large-scale avian phylogenomics [32]
CASTER	800x less CPU-intensive than two-step methods; higher accuracy under high ILS [32]	Simulation: 201 species, 10,000 loci, each 500bp [32]
STELAR	Matches ASTRAL accuracy; better than MP-EST and SuperTriplets [31]	Extensive simulations and real biological datasets [31]

Detecting Hybridization: PhyloNet in Practice

The power of network inference is illustrated in real biological studies. A phylogenomic investigation of the plant genus Lappula integrated data from 475 single-copy nuclear genes and complete plastomes [30]. The analysis revealed significant gene tree discordance. Using PhyloNet for reticulate network analysis, alongside other tools like HyDe, the researchers determined that hybridization played a crucial role in the evolution of the group. Specifically, the study found that certain clades originated via hybridization, with tetraploids arising from two independent allopolyploidization events [30]. This case study highlights how PhyloNet can be applied to unravel complex evolutionary histories involving both ILS and hybridization.

Experimental Protocols and Workflows

A Standard Phylogenomic Analysis Pipeline

A typical workflow for inferring species trees and networks involves sequential steps of data processing, gene tree estimation, and species tree/network inference.

Standard phylogenomic analysis workflow.

Detailed Protocol for Key Analyses

Protocol 1: Species Tree Inference with ASTRAL

• Input Gene Tree Collection: Obtain a set of unrooted gene trees in Newick format, estimated from multiple sequence alignments of different loci. For single-copy genes, use ASTRAL-IV; for multi-copy genes, use ASTRAL-Pro [32].
• Handling Gene Tree Uncertainty:
  • Contraction: Contract branches with low support (e.g., <10% bootstrap or <0.9 aBayes support) in the gene trees before running ASTRAL-IV to mitigate the impact of inaccurately resolved branches [32].
  • Weighting: Alternatively, use wASTRAL (weighted ASTRAL), which incorporates gene tree branch lengths and/or support values directly into the quartet score, often leading to better accuracy than contraction [32].
• Execution: Run ASTRAL with the curated set of gene trees. ASTRAL-IV's algorithm scales linearly with the number of genes, making it feasible for massive datasets [32].
• Output: The primary output is the species tree topology in Newick format. ASTRAL-IV can also compute both coalescent unit and substitution-per-site branch lengths [32].

Protocol 2: Investigating Hybridization with PhyloNet

• Prerequisite - Identifying Discordance: Before network analysis, establish that significant gene tree discordance exists that cannot be explained by ILS alone. This can be done using Quartet Sampling (QS) or metrics like the internal branch length from ASTRAL.
• Input Preparation: PhyloNet can use the same set of gene trees used for ASTRAL analysis as input, or alternatively, multiple sequence alignments.
• Inference Method Selection: Choose an appropriate inference algorithm within PhyloNet, such as Maximum Likelihood for networks or parsimony-based inference of a minimal set of hybridization events to explain the observed gene trees.
• Validation and Interpretation: Use methods like HyDe [30] to test for specific hybrid relationships and examine the support for inferred hybrid nodes. The resulting network should be interpreted in the context of known biology, such as ploidy levels and geographic distributions [30].

The Scientist's Toolkit

Table: Essential Research Reagents and Software for Phylogenomic Analysis

Tool / Resource	Function / Purpose	Application Context
IQ-TREE	Software for maximum likelihood phylogeny inference. Estimates gene trees from alignments and provides fast approximate branch supports (aBayes) [32].	Gene tree estimation for input into ASTRAL or PhyloNet.
ASTER Suite	Integrated software package for species tree inference, includes ASTRAL, wASTRAL, ASTRAL-Pro, and CASTER for different input types [32].	Primary tool for species tree inference under ILS.
PhyloNet	Software package for inference, simulation, and analysis of phylogenetic networks [30].	Modeling evolutionary histories involving hybridization and introgression.
HyDe	Software for detecting hybridization from genomic data using site pattern probabilities [30].	Testing specific hybrid hypotheses and validating PhyloNet results.
Angiosperms353 / HybPiper	Probe sets and pipelines for target sequence capture of nuclear genes from plant genomes [30].	Generating the hundreds of single-copy nuclear loci required for robust phylogenomic analysis.
Quartet Sampling (QS)	Method to quantify support and conflict in a species tree by assessing quartets of taxa [30].	Assessing the robustness of a species tree and identifying nodes with high discordance.

The comparative analysis of ASTRAL and PhyloNet reveals a complementary relationship. ASTRAL remains the gold-standard for scalable and accurate species tree inference under the MSC model, with continuous improvements in the ASTER suite enhancing its speed and versatility. When evolutionary histories are complicated by hybridization, PhyloNet provides the necessary framework to infer reticulate events. The choice between tree-based and SNP-based introgression tests is not mutually exclusive; a robust research program often integrates both. For instance, ASTRAL can establish the primary species tree, PhyloNet can identify potential hybrid nodes, and SNP-based tests like D-statistics can provide an independent signal of gene flow. As phylogenomic datasets grow in size and complexity, the combined application of these powerful tools will be essential for unraveling the intricate branches and webs of life's history.

In evolutionary genomics and drug development, accurately identifying introgressed genetic material—the transfer of genetic information between species or populations through hybridization—is critical for understanding disease mechanisms, tracing pathogen evolution, and identifying adaptive traits. The statistical detection of introgression relies primarily on two methodological frameworks: SNP-based tests and tree-based phylogenetic approaches. SNP-based methods, including the widely used ABBA-BABA test and its associated D-statistic, analyze patterns of derived alleles across populations to infer historical gene flow. In contrast, tree-based methods infer introgression by analyzing the distribution of phylogenetic tree topologies constructed from genomic sequence alignments, providing an alternative approach with different underlying assumptions [11]. Each methodology employs distinct statistical measures to quantify introgression signals, with ongoing research focused on improving their accuracy, robustness, and interpretability. Within this context, Bayesian statistics have emerged as a powerful framework for hypothesis testing, offering a principled approach to weigh evidence for competing evolutionary models. This comparative guide examines the experimental performance, underlying methodologies, and practical applications of these approaches, with particular focus on the integration of Bayesian inference through the df-BF (Bayes Factor) statistic to resolve uncertainties in introgression detection.

Methodological Frameworks: Principles and Workflows

Tree-Based Introgression Detection

Tree-based methods operate by extracting numerous sequence alignment blocks from whole-genome alignments, filtering them for quality and suitability for phylogenetic analysis, and then inferring phylogenetic trees for each block. The distribution of these tree topologies across the genome is then analyzed to detect discrepancies from the expected species tree that signal historical introgression events. This approach utilizes established phylogenetic software tools including IQ-TREE for maximum likelihood tree inference, ASTRAL for species tree estimation from gene trees, and PhyloNet for inferring species networks that explicitly model introgression events [11]. The workflow begins with whole-genome alignment data, typically in MAF (Multiple Alignment Format) format, from which suitable alignment blocks are extracted using custom Python scripts. These blocks are filtered based on completeness, number of polymorphic sites, and recombination breakpoints before phylogenetic analysis. The resulting gene trees are used to estimate a species tree and assess support for alternative phylogenetic relationships that indicate introgression [11].

SNP-Based Introgression Tests

SNP-based methods, particularly the ABBA-BABA test (D-statistic), operate on patterns of single nucleotide polymorphisms across four taxa: two sister populations (P1 and P2), an outgroup (O), and a potential introgressing population (P3). The test examines the relative frequency of two site patterns: "ABBA" sites, where P1 and O share the ancestral allele while P2 and P3 share the derived allele, and "BABA" sites, where P1 and P3 share the derived allele while P2 and O share the ancestral allele. Under no introgression, these patterns should occur with equal frequency, but significant deviations indicate asymmetric gene flow. The D-statistic quantifies this deviation as D = (∑ABBA - ∑BABA) / (∑ABBA + ∑BABA) [11]. This method assumes identical substitution rates across all species and ignores the possibility of homoplasies (multiple independent substitutions at the same site), assumptions that may be problematic when analyzing divergent species [11].

Bayesian Framework and the df-BF Statistic

Bayesian statistics provide an alternative paradigm for hypothesis testing that quantifies evidence through the Bayes Factor (BF). The BF measures the change in relative beliefs about two competing hypotheses (H0 and H1) given observed data. Mathematically, this is expressed as BF = [P(H|E)/P(H^c|E)] / [P(H)/P(H^c)] = P(E|H)/P(E|H^c), where the posterior odds equal the prior odds multiplied by the Bayes Factor [33]. In the context of introgression detection, the df-BF statistic could be formulated to compare hypotheses of introgression (H1) versus no introgression (H0) based on the distribution of tree topologies or SNP patterns. Unlike frequentist p-values, which only measure evidence against a null hypothesis, Bayes Factors quantify evidence for both null and alternative hypotheses, allow continuous monitoring of evidence as data accumulate, and incorporate prior knowledge while naturally accounting for model complexity [34]. The BF interpretation follows established guidelines, with BF10 > 3 indicating moderate evidence for the alternative hypothesis, BF10 > 10 indicating strong evidence, and values below 1/3 providing evidence for the null hypothesis [35].

Integrated Workflow for Introgression Detection

The following diagram illustrates a comprehensive workflow integrating both tree-based and SNP-based approaches with Bayesian statistical evaluation:

Comparative Performance Analysis

Methodological Comparison and Experimental Findings

Table 1: Comparative Analysis of Introgression Detection Methods

Feature	Tree-Based Methods	SNP-Based Methods (D-statistic)
Statistical Basis	Distribution of gene tree topologies [11]	Allele frequency patterns (ABBA-BABA sites) [11]
Key Assumptions	Phylogenetic models of sequence evolution	Identical substitution rates, no homoplasy [11]
Data Requirements	Whole-genome sequence alignments	Genome-wide SNP data
Computational Intensity	High (multiple phylogenetic inferences)	Moderate (pattern counting)
Handling of Divergent Species	Robust (explicit evolutionary models)	Problematic (assumptions often violated) [11]
Interpretation Framework	Posterior probabilities of introgression models	Frequentist hypothesis testing (p-values)
Software Tools	IQ-TREE, ASTRAL, PhyloNet [11]	PLINK, ADMIXTURE, custom scripts

Table 2: Empirical Performance in Case Studies

Study System	Tree-Based Results	SNP-Based Results	Bayesian Integration
Neolamprologus Cichlids [11]	Robust introgression signals despite deep divergence	Potentially misleading due to violated assumptions	PhyloNet enabled Bayesian comparison of introgression models
Pine Hybrid Zones [17]	Not explicitly reported	Asymmetric introgression favoring P. mugo ancestry	Identification of loci under selection via allele frequency spectra
Populus Trees [21]	Not explicitly reported	Not explicitly reported	Association of P. fremontii markers with 75% greater survival
Wheat Phenology [36]	Not applied	Multi-locus GWAS detected 261 trait-associated SNPs	Not explicitly reported

Bayesian Advantages in Introgression Detection

The integration of Bayesian statistics, particularly through the df-BF statistic, addresses several limitations of traditional frequentist approaches to introgression detection. Bayesian methods provide direct probability statements about hypotheses, allowing researchers to quantify evidence for both null (no introgression) and alternative (introgression present) hypotheses [34]. This contrasts with p-values, which only measure evidence against the null hypothesis and are often misinterpreted [35]. Bayesian model comparison naturally compensates for differences in model complexity through the marginal likelihood, which averages over parameter space rather than optimizing [33]. This automatic penalization of complexity protects against overfitting, a critical advantage when comparing complex evolutionary models with varying numbers of introgression events. Furthermore, Bayesian approaches allow continuous monitoring of evidence as data accumulate, without needing to adjust for multiple testing or predetermined sampling plans [34]. In practical applications, Bayesian methods have been successfully implemented in phylogenetic software such as PhyloNet, which uses Markov Chain Monte Carlo (MCMC) sampling to approximate posterior probabilities of species networks with different introgression scenarios [11].

Experimental Protocols and Implementation

Detailed Methodologies for Introgression Detection

Tree-Based Protocol

The tree-based introgression detection protocol begins with extraction of alignment blocks from a whole-genome alignment using custom Python scripts, with typical block lengths of 1,000 bp to balance phylogenetic information content against recombination probability [11]. Alignment blocks are filtered based on completeness (minimum taxon representation), proportion of parsimony-informative sites, and recombination breakpoints using methods like PhiTest. For each filtered alignment block, maximum likelihood gene trees are inferred using IQ-TREE with appropriate substitution models selected via ModelFinder. The resulting gene tree set is used to estimate a species tree under the multi-species coalescent model using ASTRAL, which accounts for incomplete lineage sorting. Introgression is detected by quantifying asymmetries in the frequencies of alternative quartet topologies around specific branches, analogous to the D-statistic but based on full sequence alignments rather than SNP patterns alone [11]. For statistical validation, PhyloNet implements Bayesian inference of species networks, sampling possible introgression scenarios with MCMC to approximate posterior probabilities of different evolutionary histories.

SNP-Based Protocol with Bayesian Enhancement

The SNP-based protocol begins with quality-controlled genomic variant data, typically applying filters for call rate (>90%), minor allele frequency (>1%), and Hardy-Weinberg equilibrium (p > 0.001) using tools like PLINK [20]. For the ABBA-BABA test, researchers identify ancestral and derived alleles using an outgroup species and count site patterns across the four-taxon structure (P1, P2, P3, O). The D-statistic is calculated as (nABBA - nBABA) / (nABBA + nBABA), with significance assessed via jackknife resampling or block bootstraping to account for linked sites [11]. To enhance this framework with Bayesian inference, the df-BF statistic can be implemented by defining hypotheses H0: D=0 (no introgression) versus H1: D≠0 (introgression present). Prior distributions for D under H1 can be established based on empirical studies of introgression effect sizes, with Cauchy or Normal priors centered at zero. The Bayes Factor is then computed as BF10 = P(Data|H1)/P(Data|H0) using numerical integration or MCMC sampling, with interpretation following established guidelines (BF10 > 3: moderate evidence for introgression; BF10 > 10: strong evidence) [35].

Research Reagent Solutions

Table 3: Essential Research Tools for Introgression Analysis

Tool/Resource	Function	Application Context
IQ-TREE [11]	Maximum likelihood phylogenetic inference	Gene tree estimation from sequence alignments
ASTRAL [11]	Species tree estimation from gene trees	Coalescent-based species tree inference
PhyloNet [11]	Phylogenetic network inference	Modeling introgression as reticulate evolution
PLINK [20]	Genome-wide association analysis & quality control	SNP data processing and filtering
JASP [35] [34]	Bayesian statistical analysis with GUI	User-friendly Bayes Factor calculation
ADMIXTURE [20]	Population structure analysis	Ancestry component estimation
Progressive Cactus [11]	Whole-genome alignment	Reference-free alignment of multiple genomes
AISNP Panels [20]	Ancestry-informative markers	Fine-scale ancestry inference in admixed populations

Discussion: Interpretation and Research Implications

Statistical Interpretation Guidelines

Interpreting the results of introgression analyses requires careful consideration of the statistical framework employed. For frequentist D-statistics, the conventional significance threshold of |D| > 0 with Z-score > 3 (equivalent to p < 0.003) is often applied, but this provides only indirect evidence against the null hypothesis of no introgression [11]. In contrast, Bayesian df-BF statistics offer more intuitive interpretation: BF10 between 1-3 provides anecdotal evidence for introgression, 3-10 moderate evidence, 10-30 strong evidence, 30-100 very strong evidence, and >100 extreme evidence for introgression [35]. Importantly, BF10 < 1/3 provides evidence for the null hypothesis of no introgression, a capability lacking in frequentist approaches. The region of practical equivalence (ROPE) can be defined around D=0 to account for biologically meaningless effect sizes, with posterior probability concentrated outside the ROPE indicating meaningful introgression [35]. Researchers should report both effect size estimates (D-statistics or posterior distributions of introgression rates) and evidence measures (p-values or Bayes Factors) to provide a complete picture, as significance alone does not indicate biological importance.

Implications for Biomedical Research

The comparative performance of introgression detection methods has significant implications for drug development and biomedical research. Tree-based approaches offer advantages when analyzing divergent pathogen strains or ancient introgression events relevant to understanding virulence evolution and drug resistance mechanisms [11]. SNP-based methods provide efficient screening for recent admixture in population genomic datasets, crucial for ensuring proper stratification in genome-wide association studies [20]. The integration of Bayesian df-BF statistics enhances both approaches by quantifying evidence strength in a directly interpretable framework, reducing false positives from multiple testing, and incorporating prior knowledge about evolutionary rates or introgression probabilities [35] [34]. In practical terms, accurate introgression detection helps identify adaptively introgressed loci that may confer disease resistance or susceptibility, trace the origin and spread of pathogenic elements across species boundaries, and understand the evolutionary history of model organisms used in drug screening [21] [17]. As genomic data generation accelerates in biomedical research, Bayesian methods provide a principled framework for evidence accumulation across studies, enabling more robust inferences about the role of introgression in disease-related traits.

The comparative analysis of tree-based and SNP-based introgression detection methods reveals complementary strengths and applications in evolutionary genomics and biomedical research. Tree-based methods offer robustness for analyzing divergent taxa and explicit modeling of evolutionary processes, while SNP-based approaches provide computational efficiency for large-scale genomic screening. The integration of Bayesian statistics through the df-BF statistic enhances both frameworks by providing direct evidence quantification, natural handling of model complexity, and incorporation of prior knowledge. As genomic datasets expand in size and complexity, Bayesian model comparison approaches will play an increasingly important role in distinguishing true biological signals from statistical artifacts, ultimately leading to more accurate inferences about evolutionary history and its biomedical implications. Researchers should select introgression detection methods based on their specific biological questions, data characteristics, and interpretive needs, while recognizing the distinct advantages offered by Bayesian statistical frameworks for quantifying evidence and comparing complex evolutionary models.

The detection of ancient introgression—the historical transfer of genetic information between species—is pivotal to understanding evolutionary dynamics. Two primary computational approaches dominate this field: tree-based methods, which use phylogenetic trees to infer evolutionary history, and SNP-based methods, which utilize patterns of single nucleotide polymorphisms. This guide objectively compares the performance of these methodologies within the context of cichlid fishes and bacterial genomes, providing experimental data and protocols to inform researchers in evolutionary genetics and genomics.

## Methodological Foundations

### Tree-Based Introgression Detection

Tree-based methods infer introgression by analyzing the distribution of phylogenetic tree topologies constructed from sequence alignments across the genome. The core premise is that introgression creates a conflict between the species tree and local gene trees [11].

Key Experimental Protocol for Tree-Based Analysis [11]:

• Data Extraction: Extract multiple sequence alignment blocks from a whole-genome alignment. For the cichlid case study, a chromosome-scale alignment of five Neolamprologus species and an outgroup (Oreochromis niloticus) was used.
• Alignment Filtering: Filter alignment blocks for phylogenetic suitability based on length (e.g., 1,000 bp), minimal missing data, and a low frequency of recombination breakpoints.
• Gene Tree Inference: For each filtered alignment block, infer a maximum-likelihood phylogenetic tree (gene tree) using software like IQ-TREE.
• Species Tree Estimation: Estimate a consensus species tree from the entire set of gene trees using a tool like ASTRAL.
• Introgression Testing: Use the distribution of gene tree topologies to test for asymmetry, which signals introgression. This can complement or validate SNP-based tests. Software like PhyloNet can be used to infer species networks that directly model introgression events.

### SNP-Based Introgression Detection

SNP-based methods detect introgression by analyzing patterns in allele frequencies and polymorphisms, often relying on summary statistics computed from genomic data [37] [38].

Key Experimental Protocol for SNP-Based Analysis (D-statistic and derivatives) [37]:

• Variant Calling: Identify biallelic SNPs from whole-genome sequencing data for four taxa with the relationship (((P1, P2), P3), O), where O is an outgroup.
• Allele Pattern Counting: For each SNP, categorize patterns based on ancestral (A) and derived (B) alleles. Key patterns are ABBA (shared derived allele between P2 and P3) and BABA (shared derived allele between P1 and P3).
• Calculate D-statistic: Compute Patterson's D, which measures the excess of one pattern over the other: D = (Σ(ABBA - BABA)) / (Σ(ABBA + BABA)). A significant deviation from zero suggests introgression between P3 and either P1 or P2.
• Apply Robust Statistics: To quantify introgression in small genomic windows and address limitations of D, use statistics like the distance fraction (df), which incorporates pairwise genetic distances (dxy), or fd [37]. Another alternative is RNDmin, which uses minimum pairwise sequence distance normalized by divergence to an outgroup to detect even rare introgressed lineages [38].

## Performance Comparison: Tree-Based vs. SNP-Based Methods

The table below summarizes the comparative performance of tree-based and SNP-based methods based on analyses of simulated and empirical datasets.

Table 1: Comparative performance of tree-based and SNP-based introgression detection methods

Feature	Tree-Based Methods	SNP-Based Methods (D-statistic)
Underlying Principle	Comparison of gene tree topologies to a species tree [11]	Analysis of allele patterns (ABBA/BABA) across taxa [37]
Key Strength	More robust when analyzing divergent species with different evolutionary rates or homoplasy [11]	High power and straightforward interpretation in closely related species with even evolutionary rates [37] [38]
Key Weakness	Computationally intensive; requires multiple high-quality sequence alignments [11]	Assumes identical substitution rates and no homoplasy; can produce false positives in regions of low recombination/divergence [11] [37]
Power to Detect Rare Introgression	Limited if the introgressed lineage is not sampled or does not create a distinct topology	RNDmin and related statistics (dmin, Gmin) are sensitive to rare, recent migrants [38]
Robustness to Mutation Rate Variation	Inherently accounts for variation through tree branch lengths	Requires specific statistics (e.g., RNDmin, Gmin) to be robust; standard D is not [38]
Data Requirements	Genome-wide sequence alignments or sets of orthologous genes [11]	Genome-wide SNP data from at least three ingroup taxa and an outgroup [37]
Quantification of Introgression	Model-based approaches in PhyloNet can estimate proportions of introgression [11]	Statistics like fd and df are designed to quantify the proportion of introgression [37]

## Case Study: Introgression in Cichlid Fishes

Application of these methods to cichlid fishes has revealed the pervasive nature of introgression.

### Princess Cichlids (Neolamprologus)

A genomic study on Princess cichlids from Lake Tanganyika used whole-genome sequencing and phylogenomic analyses. It found evidence for multiple introgression events affecting different stages of diversification. A key finding was that the genomic landscape of introgression is heterogeneous: chromosome centers, with low recombination, showed less introgression and are potential reservoirs of incompatibility genes, while chromosome peripheries, with high recombination, were more dynamic and prone to adaptive introgression [39].

### Peacock Cichlids (Cichla)

A multi-locus study on Amazonian peacock cichlids used mtDNA, nuclear sequences, and microsatellites to delimit species and quantify introgression. The study highlighted that the estimated frequency of hybrid individuals is highly dependent on the species concept applied. Under a polytypic species concept (PTSC), about 2% of individuals showed hybrid ancestry, whereas under a diagnostic species concept (DSC), this figure rose to ~12%. Regardless of the concept, a significant majority of the delimited species (60-75%) showed evidence of introgression from at least one other species, including between non-sister lineages from different major clades [40]. This demonstrates that introgression is a widespread and natural, though often ephemeral, part of cichlid evolution.

## Essential Research Reagents and Computational Tools

Successful implementation of the protocols requires a suite of specialized software tools.

Table 2: Key research reagents and software solutions for introgression detection

Tool Name	Type/Function	Brief Description
IQ-TREE [11]	Phylogenetic Inference	Efficient software for maximum likelihood estimation of phylogenetic trees from molecular sequences.
ASTRAL [11]	Species Tree Estimation	Accurately estimates species trees from a set of gene trees, accounting for incomplete lineage sorting.
PhyloNet [11]	Phylogenetic Network Inference	Infers species trees and networks that can explicitly model introgression and other reticulate evolutionary events.
PopGenome [37]	Population Genomic Analysis	An R package for population genetic analyses, including the calculation of D, df, and fd statistics.
PAUP* [11]	Phylogenetic Analysis	A general-utility program for phylogenetic inference under parsimony, likelihood, and distance criteria.
FigTree [11]	Tree Visualization	Graphical viewer for phylogenetic trees, enabling visualization and annotation of tree-based results.
PhyloSNP [41]	SNP-based Phylogenetics	Builds phylogenetic trees directly from whole-genome SNP/SNV profiles, useful for bacterial and viral genomes.

## Workflow Visualization

The following diagrams illustrate the logical workflows for tree-based and SNP-based introgression detection.

### Tree-Based Introgression Workflow

### SNP-Based Introgression Workflow

The choice between tree-based and SNP-based methods for detecting ancient introgression is not a matter of one being universally superior. Instead, the optimal approach depends on the biological question, the divergence time of the taxa, and the nature of the available data. Tree-based methods offer robustness in complex scenarios involving divergent species and heterogeneous genomic landscapes, as demonstrated in cichlid studies [11] [39]. SNP-based methods provide powerful and efficient detection in closely related species and are particularly adept at identifying recent and rare introgression events [37] [38]. A synergistic approach, where signals from both methodologies are integrated and validated, is often the most effective strategy for uncovering the complex history of introgression shaping genomes.

Navigating Pitfalls and Enhancing Accuracy in Introgression Analysis

Detecting introgression—the exchange of genetic material between species through hybridization and backcrossing—is fundamental to understanding evolutionary dynamics. While single nucleotide polymorphism (SNP)-based methods have become standard tools for this purpose, they face a significant challenge: evolutionary rate variation between lineages can generate false-positive signals that mimic genuine introgression. This problem becomes increasingly severe as the divergence time between studied taxa increases, potentially leading to incorrect conclusions about evolutionary history [42].

The core issue lies in the fundamental assumptions underlying popular SNP-based tests like the D-statistic (ABBA-BABA test). These methods assume uniform evolutionary rates across lineages and minimal homoplasy (independent substitutions at the same genomic site). However, in reality, factors such as generation time, metabolic rate, and environmental pressures create substantial rate variation across lineages. When combined with the increasing probability of homoplasy over longer evolutionary timescales, these violations of method assumptions create systematic patterns that can be misinterpreted as evidence of introgression [42].

This article provides a comparative analysis of SNP-based and tree-based methods for introgression detection, with particular focus on how evolutionary rate variation affects their performance. We present experimental data quantifying false-positive rates, detail methodological protocols for both approaches, and provide recommendations for researchers seeking reliable introgression inference across diverse evolutionary timescales.

Mechanisms of Misleading Signals: How Rate Variation Creates False Positives

Theoretical Foundations of the Problem

The D-statistic operates by comparing patterns of ancestral ("A") and derived ("B") alleles across four taxa. In the absence of introgression but presence of incomplete lineage sorting (ILS), two sister species are expected to share equal proportions of derived alleles with a third, outgroup species. A statistically significant imbalance in ABBA versus BABA site patterns is interpreted as evidence of introgression [42].

Evolutionary rate variation disrupts this expectation through a specific mechanism: lineages with accelerated substitution rates accumulate more homoplasies—independent mutations at identical sites—which are more likely to produce convergent allele patterns that mimic introgression signals. As illustrated in Figure 1, when different evolutionary rates create heterogenous branch lengths, the probability of homoplasy increases substantially in faster-evolving lineages, generating false signals of introgression that are statistically significant but biologically misleading [42].

Empirical Evidence of the Problem

Simulation studies demonstrate the severity of this effect. Under conditions of divergent evolutionary rates between lineages, the D-statistic can produce false-positive rates exceeding acceptable thresholds—particularly when analyzing deeply divergent taxa. One comprehensive simulation analysis found that "some commonly applied statistical methods, including the D-statistic and certain tests based on sets of local phylogenetic trees, can produce false-positive signals of introgression between divergent taxa that have different rates of evolution" [42]. These misleading signals become increasingly pronounced with greater degrees of rate variation and deeper phylogenetic divergences.

Table 1: Factors Increasing False-Positive Risk in SNP-Based Introgression Detection

Factor	Effect on False-Positive Risk	Underlying Mechanism
Increasing divergence time	Substantial increase	Higher probability of homoplasious substitutions
Greater rate variation	Substantial increase	Accelerated homoplasy in fast-evolving lineages
Complex population structure	Moderate increase	Ancestral structure mimics introgression patterns
Poor reference genome quality	Moderate increase	Misalignment creates artificial SNP patterns
Relaxed mapping stringency	Moderate increase	Increased mismapping artifacts

Comparative Performance: SNP-Based vs. Tree-Based Methods

Quantitative Performance Comparison

Different methodological approaches show substantially varying susceptibility to false positives caused by evolutionary rate variation. While SNP-based methods like the D-statistic are highly vulnerable to this effect, tree-based alternatives demonstrate greater robustness, particularly for deeper divergences [42].

Table 2: Performance Comparison of Introgression Detection Methods Under Rate Variation

Method	Core Approach	False-Positive Rate with Rate Variation	Optimal Application Context
D-statistic	SNP allele frequency patterns	High (particularly for deep divergences)	Recent divergence, minimal rate variation
Tree-based D-statistic (Dtree)	Local tree topology frequencies	Moderate	Moderate to deep divergences
Random Forests	Machine learning with decision trees	Low to moderate	Complex architectures, epistatic interactions
Logic Regression	Boolean combinations of SNPs	Low to moderate	Pathway-based analyses, specific interactions
Dsuite	Clustering of introgressed sites	Low	Specifically designed for rate variation contexts

Tree-based methods generally demonstrate superior performance under conditions of rate variation because they operate on phylogenetic topologies inferred from longer genomic segments rather than individual SNP patterns. This provides a buffer against homoplasy, as convergent single mutations rarely affect overall tree topology, whereas concerted homoplasy patterns across multiple sites are less probable [42]. One tree-based approach, the tree-based D-statistic (Dtree), analyzes frequencies of different local tree topologies, with significant imbalances in alternative topologies suggesting introgression. While more robust than SNP-based D-statistics, Dtree can still produce false positives under extreme rate variation [42].

Type I Error Control in Association Mapping

The challenge of false positives extends beyond introgression detection to association mapping. Studies comparing tree-based and non-tree-based association mapping methods have revealed important differences in type I error control. In one investigation, a non-tree-based t-test showed type I error rates above 0.05 across all five genes studied, while a tree-based likelihood score statistic (LSS) approach consistently maintained error rates below 0.05, demonstrating more conservative and reliable behavior [6].

Experimental Protocols for Introgression Detection

Protocol for SNP-Based Introgression Detection

The following protocol outlines standard procedures for SNP-based introgression detection using the D-statistic:

Step 1: Data Preparation and Quality Control

• Assemble whole-genome sequencing data for at least four taxa (P1, P2, P3, and outgroup)
• Perform quality filtering of raw reads (remove adapters, low-quality bases)
• Map reads to a reference genome using tools like BWA or Bowtie2 [43]
• Apply strict mapping quality filters (MAPQ ≥ 20) to reduce mismapping [44]
• Remove PCR duplicates and perform local realignment around indels

Step 2: Variant Calling and Filtering

• Call variants using specialized tools (GATK, FreeBayes, or SAMtools) [43]
• Apply base quality score recalibration (GATK) or equivalent quality control
• Filter SNPs based on depth coverage (e.g., minimum depth = 5, maximum depth = 150) [44]
• Retain only biallelic SNPs with quality scores ≥ 20

Step 3: D-Statistic Calculation

• Polarize alleles using the outgroup sequence
• Identify ABBA and BABA sites across the genome
• Calculate the D-statistic: D = (ABBA - BABA) / (ABBA + BABA)
• Assess statistical significance using block jackknifing or permutation tests

Critical Considerations: This approach is most reliable when analyzing closely related species with similar evolutionary rates. For deeper divergences, additional validation through tree-based methods is strongly recommended [42].

Protocol for Tree-Based Introgression Detection

The following protocol details tree-based introgression detection using the Dtree approach:

Step 1: Data Preparation and Alignment

• Follow identical data preparation steps as for SNP-based protocol
• Partition genome into non-overlapping windows (e.g., 1-10 kb depending on divergence)
• Align sequences for each window across all taxa

Step 2: Gene Tree Inference

• Infer phylogenetic trees for each genomic window using maximum likelihood (RAxML, IQ-TREE) or Bayesian methods (BEAST2)
• Model rate variation among lineages using appropriate clock models (relaxed lognormal)
• Assess node support with bootstrapping or posterior probabilities

Step 3: Dtree Calculation

• For each four-taxon set (P1, P2, P3, outgroup), categorize local trees by topology
• Count frequencies of the major topology (ABBA analog) and secondary topology (BABA analog)
• Calculate Dtree = (Freqmajor - Freqsecondary) / (Freqmajor + Freqsecondary)
• Assess statistical significance with appropriate multiple testing correction

Critical Considerations: Tree-based methods require sufficient phylogenetic signal in each genomic window. Window size should be optimized to ensure reliable tree reconstruction while maintaining adequate genomic resolution [42].

Figure 1: Comparative Workflow for SNP-Based and Tree-Based Introgression Detection

The Researcher's Toolkit: Essential Methods and Software

Table 3: Research Reagent Solutions for Introgression Analysis

Tool Category	Specific Software	Primary Function	Key Considerations
Read Mappers	BWA, Bowtie2	Alignment of sequencing reads to reference	Mapping accuracy critical; avoid overly relaxed mismatch settings [44]
Variant Callers	GATK, FreeBayes, SAMtools	Identification of SNPs from aligned reads	FreeBayes uses Bayesian models; GATK employs de novo assembly [43]
Tree Inference	RAxML, IQ-TREE, BEAST2	Phylogenetic tree reconstruction for genomic regions	Model selection important for handling rate variation [42]
Introgression Tests	Dsuite, ADMIXTOOLS	Implementation of D-statistic and related tests	Dsuite includes specific tests for rate variation contexts [42]
Population Structure	ADMIXTURE, PLINK	Ancestral component analysis	Useful for validating introgression signals [45]

The challenge of false positives in SNP-based introgression detection presents a significant methodological concern, particularly for studies of deeply divergent taxa. Evolutionary rate variation between lineages systematically generates patterns that mimic genuine introgression, potentially leading to incorrect evolutionary inferences [42].

Based on comparative performance data, we recommend:

• Method Selection Based on Divergence Time: For recently diverged taxa (<1-2 million years), SNP-based methods like the D-statistic remain appropriate. For deeper divergences, tree-based approaches offer greater reliability.

• Validation Through Multiple Approaches: Significant signals from SNP-based tests should be validated using tree-based methods, particularly when analyzing taxa with suspected rate variation.

• Careful Parameterization: Regardless of method, strict quality control, appropriate filtering thresholds, and careful model selection are essential for minimizing false positives.

• Acknowledgment of Limitations: Researchers should explicitly acknowledge and account for the limitations of their chosen methods when drawing evolutionary conclusions.

As genomic datasets continue to grow in size and taxonomic breadth, the development of more robust methods for introgression detection—particularly those specifically designed to handle evolutionary rate variation—represents an important frontier in evolutionary genomics.

The Impact of Homoplasy and Recombination on Method Performance

Phylogenetic analysis aims to reconstruct evolutionary histories, but this process is often complicated by evolutionary forces that obscure true genealogical relationships. Homoplasy—the independent emergence of identical genetic variants in distinct lineages—and recombination—the exchange of genetic material between lineages—represent two such significant challenges. Homoplasy can arise from parallel evolution, convergent evolution, or evolutionary reversions, creating patterns that mimic shared ancestry [46]. Recombination, particularly widespread in bacterial species, results in different genomic regions following distinct phylogenetic histories [47] [48].

These forces directly impact the performance and reliability of phylogenetic methods. Tree-based approaches that assume a single evolutionary history for entire genomes struggle with recombined regions, while SNP-based methods for detecting introgression can produce false positives when evolutionary rate variation generates homoplastic sites [49] [48]. This guide provides an objective comparison of method performance under these challenges, equipping researchers with evidence-based selection criteria for their phylogenetic analyses.

Performance Comparison of Phylogenetic Methods

Quantitative Comparison of Method Performance

Table 1: Performance comparison of introgression detection methods under rate variation

Method	Core Principle	Key Assumption	False Positive Rate with Moderate Rate Variation	Sensitivity to Shallow Phylogenies	Computational Demand
D-statistic	ABBA-BABA site pattern asymmetry	No multiple hits (single mutation per site)	Up to 100% with 33% rate variation [49]	High sensitivity in young phylogenies [49]	Low
HyDe	Site pattern frequency comparison	Molecular clock among lineages	Similar to D-statistic [49]	High sensitivity in young phylogenies [49]	Low to Moderate
SNPPar	Ancestral state reconstruction	Accurate reference genome and tree	High specificity (zero false positives in simulations) [46]	Not specifically evaluated	Moderate
ptACR	Site compatibility with permutation testing	Phylogenetic incongruence indicates recombination	Lower false positive rate than basic ACR [47]	Effective across timescales	Moderate

Table 2: Performance of species tree inference methods with SNP data

Method	Approach	Tolerance to Missing Data	Handling of Homoplasy/Recombination	Topological Accuracy with Patchy Data
SNAPP	Bayesian coalescent	Low [50]	Not specifically designed for recombination	Fails with large missing data [50]
SVDquartets	Coalescent-based quartet analysis	Moderate [50]	Not specifically designed for recombination	Correct topology with complete data [50]
Allele-wise Bayesian	Species-level allele frequency summary	High [50]	Limited inherent protection	Good approximation to SNAPP [50]
Dollo Parsimony	Presence/absence of derived alleles	High [50]	Some protection through character weighting	Congruent with SNAPP for empirical data [50]

Impact of Evolutionary Forces on Method Performance

Homoplasy and Rate Variation Effects

Even minor deviations from a molecular clock can severely impact site-pattern methods. In shallow phylogenies (approximately 3×10⁵ generations) with small population sizes, weak rate variation (17% difference) between sister lineages can inflate false positive rates for introgression up to 35% using a 500 Mb genome dataset. Moderate rate variation (33% difference) can increase false positive rates to 100% under the same conditions [49]. This occurs because rate heterogeneity creates asymmetries in ABBA and BABA site patterns that mimic introgression signals. The problem intensifies when using more distant outgroups, which further amplifies these spurious signals [49].

Homoplasy impacts different methods variably. The D-statistic's underlying assumption of no multiple hits makes it particularly vulnerable to homoplasy, as homoplastic sites can create ABBA-BABA asymmetry that mimics introgression [49]. In contrast, SNPPar demonstrates high specificity in homoplasy identification, showing zero false positives across all tests with simulated Mycobacterium tuberculosis data while maintaining high sensitivity (zero false-negatives in 89% of tests) [46].

Recombination Effects

The pervasive nature of recombination in bacterial genomes fundamentally challenges tree-based phylogenetic approaches. For most bacterial species, each genomic locus has been overwritten by recombination many times, with phylogenies changing thousands of times along the genome [48]. In Escherichia coli, the majority of genomic differences between strains result from recombination events rather than clonal inheritance, with most strain pairs sharing no DNA from their clonal ancestor [48].

Compatibility-based methods like ptACR offer advantages for recombination detection by identifying phylogenetic incongruence without requiring tree reconstruction. The permutation test approach in ptACR reduces false positive rates compared to basic ACR while maintaining similar sensitivity, effectively identifying recombination breakpoints in bacterial pathogens like Staphylococcus aureus [47].

Experimental Protocols and Methodologies

Assessing Introgression Method Robustness to Rate Variation

Objective: To evaluate the false positive rates of D-statistic and HyDe under controlled rate variation conditions [49].

Workflow:

• Theoretical Analysis: Derive expected D-values mathematically under varying rate variation, phylogenetic age, population size, and outgroup distance parameters
• Simulation Framework: Simulate a range of evolutionary scenarios spanning 10⁴ to 10⁶ generations using the multispecies coalescent with introgression (MSci) model
• Parameter Manipulation: Systematically introduce rate variation between sister lineages (0% to 33% difference)
• Method Application: Apply D-statistic and HyDe to simulated datasets with known evolutionary histories (no introgression)
• False Positive Quantification: Calculate false positive rates as the proportion of simulations incorrectly detecting significant introgression

Key Metrics: False positive rate, D-statistic significance, HyDe significance, power analysis [49].

Homoplasy Identification Protocol

Objective: To identify homoplasic SNPs and classify them by type (parallel, convergent, or revertant) [46].

Workflow:

• Input Preparation:
  • SNP alignment from whole genome sequencing
  • Reference tree (from RAxML-NG or IQ-TREE)
  • Annotated reference genome
• Ancestral State Reconstruction: Use TreeTime for maximum likelihood inference of ancestral states
• Mutation Mapping: Assign mutation events to specific tree branches using monophyly tests
• Homoplasy Identification: Identify SNPs where the same derived nucleotide emerges independently in multiple lineages
• Classification:
  • Parallel: Same substitution (e.g., A→T) independently in multiple lineages
  • Convergent: Same nucleotide via distinct substitution pathways (e.g., A→T in one lineage, A→G→T in another)
  • Revertant: Derived nucleotide reverting to ancestral state
• Annotation: Annotate effects at codon and gene levels to identify convergent evolution

Key Metrics: Specificity, sensitivity, classification accuracy, computational efficiency [46].

Recombination Breakpoint Detection

Objective: To identify statistically significant recombination breakpoints in bacterial genomes [47].

Workflow:

• Compatibility Calculation: For each informative site in a multiple sequence alignment, calculate pairwise compatibility scores within a sliding window (default size: 200 sites)
• Average Compatibility Ratio: Compute the average of all pairwise compatibility scores within the window
• Local Minima Identification: Identify sites with local minima in average compatibility ratio as potential breakpoints
• Permutation Testing:
  • Randomly shuffle sites within the window 10,000 times
  • Generate null distribution of compatibility statistics
  • Calculate p-values as the proportion of permuted statistics less than or equal to the observed value
• Multiple Testing Correction: Apply false discovery rate control to identify statistically significant breakpoints

Key Metrics: False positive rate, sensitivity, F1 score, breakpoint accuracy [47].

Workflow Visualization

The Scientist's Toolkit

Table 3: Essential research reagents and computational tools

Tool/Reagent	Primary Function	Application Context	Key Considerations
TreeTime [46]	Ancestral state reconstruction and homoplasy identification	Phylogenetic analysis of large genomic datasets	Linear execution time with sample size; requires pre-calculated tree
ptACR [47]	Recombination breakpoint detection with statistical testing	Bacterial genome evolution studies	Compatibility-based; more efficient than phylogenetic methods
SNPPar [46]	Homoplasic SNP detection and classification	Adaptive evolution studies in pathogens	High specificity; efficient with large datasets (>1000 isolates)
SNAPP [50]	Bayesian species tree inference from SNP data	Species delimitation and shallow phylogenetics	Computationally demanding; low tolerance for missing data
SVDquartets [50]	Coalescent-based species tree estimation	Species tree inference with patchy data	Moderate missing data tolerance; fails with extensive missing data
ClonalFrameML [47]	Recombination detection and ancestral reconstruction	Bacterial phylogenetics	Maximum likelihood approach; uses hidden Markov model
D-statistic [49]	Introgression detection using site patterns	Hybridization and gene flow studies	Highly sensitive to rate variation; false positives under molecular clock violation
SIMCOAL 2 [50]	Coalescent simulation of SNP data	Method validation and testing	Customizable demographic scenarios; generates synthetic datasets with known truth

Homoplasy and recombination present distinct challenges that differentially affect phylogenetic method performance. Site-pattern methods like D-statistic and HyDe show extreme sensitivity to rate variation—even moderate (33%) differences can produce 100% false positive rates for introgression detection [49]. For recombination-heavy datasets like bacterial genomes, compatibility methods (ptACR) provide more reliable breakpoint identification with lower false positive rates [47]. When homoplasy is the primary concern, SNPPar offers exceptional specificity for identifying genuinely homoplasic sites [46].

Method selection must be guided by dataset properties and evolutionary context. For species tree inference from SNP data with missing values, SVDquartets and allele-wise Bayesian approaches provide reasonable alternatives when SNAPP is computationally prohibitive [50]. Critically, researchers should verify that their chosen methods' assumptions align with their biological system's evolutionary dynamics, particularly regarding rate variation, recombination frequency, and phylogenetic depth.

The accurate detection of introgressed genomic regions—where genetic material has been transferred between species or populations—heavily depends on the quality of input data. In comparative genomic studies, the initial alignment blocks extracted from whole-genome alignments often contain varying degrees of missing data, sequencing errors, and recombinant regions that can severely bias phylogenetic inference and introgression signals. The crucial process of filtering these alignment blocks and managing missing data represents a fundamental divergence between tree-based and SNP-based introgression detection methods, with significant implications for their respective performances under different evolutionary scenarios.

SNP-based methods like the ABBA-BABA test (D-statistic) assume identical substitution rates across all species and the absence of homoplasies (multiple independent substitutions at the same site), conditions that likely hold for recently diverged species but become problematic when analyzing more divergent taxa [11]. In contrast, phylogenetic approaches based on sequence alignments can incorporate more complex evolutionary models, potentially offering verification or rejection of patterns identified through SNP-based methods [11]. This methodological comparison frames the critical importance of optimized data filtering protocols, as the susceptibility of each approach to data quality issues varies substantially.

Methodological Comparison: Data Handling in Tree-Based vs. SNP-Based Approaches

Fundamental Differences in Data Processing

Table 1: Core Methodological Differences in Data Handling

Aspect	Tree-Based Methods	SNP-Based Methods
Primary Data Unit	Sequence alignment blocks (contiguous regions)	Individual SNPs (single nucleotides)
Missing Data Handling	Filtering of incomplete alignment blocks; potential for explicit modeling in phylogenetic inference	Typically implemented through individual filtering; may exclude sites with missing calls
Recombination Handling	Explicit detection and filtering of recombinant alignment blocks	Often assumed to be minimal or accounted for via SNP pruning
Evolutionary Models	Incorporates complex substitution models accounting for multiple hits, rate variation	Generally assumes no homoplasy and constant substitution rates
Key Assumptions	Models can accommodate rate variation and some homoplasy	Assumes identical substitution rates and minimal homoplasy [11]
Optimal Taxonomic Scope	More divergent species with complex evolutionary histories	Recently diverged species where key assumptions hold [11]

Alignment Block Filtering Criteria and Thresholds

Table 2: Quantitative Filtering Criteria for Alignment Blocks

Filtering Parameter	Threshold	Rationale	Impact on Downstream Analysis
Alignment Block Length	Minimum 1,000 bp [11]	Balance between information content and recombination probability	Shorter blocks reduce phylogenetic signal; longer blocks increase recombination risk
Taxon Completeness	Ideally 100% species representation per block	Ensures comprehensive phylogenetic representation	Missing taxa create incomplete gene trees, reducing phylogenetic resolution
Proportion of Missing Data	Variable; optimize based on empirical distributions	Maximizes informative sites while retaining sufficient data	Excessive filtering reduces dataset size and statistical power
Recombination Signal	Remove alignments with strongest signals [11]	Prevents phylogenetic inaccuracy from conflated histories	Reduces topological inconsistencies in gene tree estimation
Polymorphic Sites	Context-dependent; retain informative but not overly divergent loci	Ensures sufficient phylogenetic signal while minimizing saturation	Balances signal-to-noise ratio in tree inference

Experimental Protocols for Data Quality Optimization

Alignment Block Extraction and Filtering Workflow

The following protocol for extracting and filtering alignment blocks from whole-genome alignments is adapted from established phylogenetic introgression detection pipelines [11]:

Step 1: Extract Alignment Blocks from Whole-Genome Alignment

• Input: Whole-genome alignment in MAF (Multiple Alignment Format) or similar format
• Process: Use customized Python scripts or bioinformatics tools (e.g., HAL tools) to extract contiguous alignment blocks
• Parameters: Define minimum block length (typically 1,000 bp as starting point) [11]
• Output: Set of candidate alignment blocks for further filtering

Step 2: Filter by Taxon Completeness

• Retain only alignment blocks containing sequences for all target species
• In cases of incomplete datasets, establish minimum taxon representation threshold (e.g., ≥80% of species)
• Exclude blocks with patchy species representation that would yield incomplete gene trees

Step 3: Assess and Filter by Missing Data Proportion

• Calculate proportion of missing data (gaps or ambiguous bases) per alignment
• Establish empirical threshold based on distribution of completeness across all blocks
• Retain blocks with missing data below established threshold (e.g., ≤20% missing data)

Step 4: Quantify and Filter by Recombination Signals

• Implement recombination detection algorithms (e.g., PhiTest, GARD) to identify breakpoints
• Quantify recombination signals per alignment using appropriate statistics
• Remove alignments with strongest recombination signals to minimize phylogenetic inaccuracy [11]

Step 5: Assess Information Content

• Calculate number of parsimony-informative sites or phylogenetic informativeness
• Retain blocks with sufficient information content for tree inference
• Balance between conservative filtering (minimizing noise) and retaining adequate dataset size

Workflow Visualization

Comparative Performance Assessment Protocol

To objectively compare the impact of data filtering on tree-based versus SNP-based introgression detection, implement the following experimental design:

Experimental Setup:

• Dataset Preparation: Use a validated whole-genome alignment with known introgression events (e.g., cichlid fish dataset [11])
• Filtering Treatments: Apply graduated filtering stringency:
  • Minimal filtering (length >500 bp only)
  • Moderate filtering (length >1,000 bp, missing data <30%)
  • Stringent filtering (length >1,000 bp, missing data <10%, recombination filtering)
• Method Application: Apply both tree-based and SNP-based methods to each filtered dataset
• Performance Metrics: Compare methods using:
  • Consistency with known introgression events
  • Statistical support values (bootstrap support for trees; Z-scores for D-statistics)
  • Concordance across methods

Analysis Implementation:

• For tree-based approaches: Generate maximum-likelihood gene trees using IQ-TREE [11], infer species tree with ASTRAL [11], assess topological asymmetry
• For SNP-based approaches: Calculate D-statistics (ABBA-BABA tests) with appropriate outgroup selection
• Cross-validate results between methods to identify consistent signals

The Scientist's Toolkit: Essential Research Reagents and Computational Tools

Table 3: Research Reagent Solutions for Introgression Analysis

Tool/Category	Specific Implementation	Function in Analysis
Whole-Genome Alignment	Progressive Cactus [11]	Reference-free alignment of multiple genomes
Alignment Processing	HAL tools, custom Python scripts	Conversion between alignment formats; block extraction
Phylogenetic Inference	IQ-TREE v.2 [11], PAUP* [11]	Maximum likelihood tree estimation from sequence alignments
Species Tree Estimation	ASTRAL [11]	Coalescent-based species tree from gene trees
Recombination Detection	PhiTest, GARD	Identification of recombination breakpoints in alignments
Introgression Tests	D-statistics (ABBA-BABA), PhyloNet [11]	Detection of gene flow between lineages
Visualization	FigTree [11]	Visualization and manipulation of phylogenetic trees
Population Genomic Analysis	ADMIXTURE [20], PLINK [20]	Ancestry component estimation; genotype data processing

Comparative Performance Data: Tree-Based vs. SNP-Based Methods

Performance Under Different Data Quality Scenarios

Table 4: Method Performance Across Filtering Stringency Levels

Data Quality Scenario	Tree-Based Method Performance	SNP-Based Method Performance	Key Observations
Minimally Filtered Data	Moderate accuracy; susceptible to recombination artifacts	High false-positive rate; violates key assumptions	Both methods show reduced reliability with poor data quality
Moderately Filtered Data	High accuracy; robust to moderate missing data	Improved accuracy with recent divergence	Tree methods show advantage for divergent taxa [11]
Stringently Filtered Data	Maximum accuracy; potential for reduced statistical power due to fewer loci	Optimal for recent divergence; may lack power for deep divergence	Data loss from overt filtering affects both methods
High Missing Data (>30%)	Resilient with appropriate model specification	Severely compromised; incomplete site patterns	Tree methods superior for incomplete datasets
Strong Recombination Signals	Compromised unless properly filtered	Violates phylogenetic independence assumptions	Highlights critical need for recombination filtering [11]

Empirical Performance Evidence

Recent systematic analyses across diverse taxonomic groups provide empirical evidence for methodological performance:

In bacterial genomics, where introgression detection faces unique challenges, tree-based approaches identified an average of 2% introgressed core genes across 50 major lineages, with up to 14% introgression in Escherichia-Shigella [51]. These estimates, however, were highly dependent on accurate species delimitation and filtering of ambiguous regions, highlighting the critical importance of data quality control steps.

Plant genomic studies on Chinese wingnuts (Pterocarya species) demonstrated that tree-based methods successfully identified introgressed regions containing candidate genes for environmental adaptation (TPLC2, CYCH;1, LUH, bHLH112) [52]. These regions showed lower genetic load and higher genetic diversity compared to the genomic background, providing biological validation of the introgression signals detected through phylogenetic approaches.

Vertebrate studies on pufferfish (Takifugu) genomes revealed that introgression detection played a crucial role in understanding speciation mechanisms, particularly for T. niphobles and T. oblongus [53]. The integration of tree-based methods with population genomic approaches provided strong evidence for introgression-driven speciation, validated through multiple independent lines of evidence.

Integrated Analysis Framework and Decision Pathway

To guide researchers in selecting and applying appropriate filtering strategies for their specific research context, the following decision pathway incorporates both methodological considerations and taxonomic scope:

The comparative performance of tree-based versus SNP-based introgression detection methods is inextricably linked to data quality optimization through appropriate filtering of alignment blocks and management of missing data. While SNP-based methods (e.g., D-statistics) offer computational efficiency and straightforward interpretation for recently diverged taxa with high-quality data, tree-based approaches provide greater robustness for analyzing divergent lineages and datasets with complex evolutionary histories.

Strategic filtering protocols that balance the competing demands of data completeness and quality control are essential for accurate introg inference. The empirical evidence across diverse taxonomic groups—from bacteria and plants to vertebrates—consistently demonstrates that method performance depends critically on appropriate data handling tailored to specific evolutionary contexts. By implementing the systematic filtering workflows and comparative frameworks outlined here, researchers can significantly enhance the reliability of introgression detection across the tree of life.

The detection of introgression—the transfer of genetic material between species or populations through hybridization—has been revolutionized by genomic data and sophisticated computational methods. Choosing the correct inference strategy is paramount for evolutionary biologists, as the effectiveness of different tests varies significantly based on the divergence time between taxa and the type of genetic data available [54]. This guide provides an objective comparison of two fundamental approaches: tree-based methods rooted in the multispecies coalescent (MSC) framework and SNP-based methods often coupled with machine learning. The selection between these strategies carries substantial implications for accurately reconstructing evolutionary histories, understanding adaptive processes, and correctly identifying gene flow patterns that shape biodiversity. As genomic datasets expand in both size and complexity, a systematic framework for method selection becomes increasingly essential for researchers across evolutionary biology, conservation genetics, and forensics.

Core Concepts: Tree-Based vs. SNP-Based Approaches

Tree-based methods operate within the multispecies coalescent framework, explicitly modeling gene tree histories within a species tree or network. These methods treat gene flow as a fundamental parameter of the evolutionary model. The two primary models are the MSC-with-Introgression (MSC-I), which models gene flow as discrete pulses at specific time points, and the MSC-with-Migration (MSC-M), which models continuous gene flow at a constant rate over time [54]. These full-likelihood methods use complete sequence information from multiple loci, accommodating incomplete lineage sorting and providing a robust statistical foundation for parameter estimation.

SNP-based methods typically utilize ancestry-informative single nucleotide polymorphisms (AISNPs) analyzed through population genetic or machine learning approaches. Rather than modeling the complete genealogical process, these methods often focus on allele frequency differences, ancestry components, or geographic patterns. Recent advances combine carefully designed AISNP panels with machine learning algorithms—including logistic regression, support vector machines, random forests, and convolutional neural networks—to classify genetic ancestry or predict geographic origins [20]. The Locator framework exemplifies this approach, using deep neural networks to predict latitude and longitude directly from unphased genotypes [20].

The fundamental distinction lies in their treatment of genetic data: tree-based methods model the genealogical process underlying sequence data, while SNP-based methods typically operate on patterns within genotypic data, often with sophisticated computational approaches rather than explicit evolutionary models.

Comparative Performance Analysis

Method Performance Across Scenarios

Table 1: Comparative Performance of Introgression Detection Methods

Method Category	Specific Method/Model	Optimal Divergence Context	Data Requirements	Key Performance Metrics	Detects Direction of Gene Flow?
Tree-Based (MSC-I)	BPP, Phylonet	Deep to moderate divergence	Sequence data from multiple loci (UCEs, AHE, exomes)	High accuracy for recent pulses; struggles with continuous migration [54]	Yes, with correct model specification [54]
Tree-Based (MSC-M)	BPP, *BEAST	Moderate to recent divergence with ongoing gene flow	Sequence data from multiple loci	Effective for continuous migration; computationally intensive [54]	Yes, but misspecification causes biases [54]
SNP-Based (Population Genetic)	ADMIXTURE, f-statistics	Various divergence times	Genome-wide SNP data (e.g., AISNPs)	Limited by reference populations; model assumptions [20]	Most methods cannot identify direction [54]
SNP-Based (Machine Learning)	XGBoost, Locator	Fine-scale population structure	50-2,000 AISNPs [20]	95.6% accuracy with 2,000 AISNPs; AUC=0.999 [20]	Not primary focus; excels at classification [20]

Quantitative Performance Data

Table 2: Empirical Performance Metrics from Published Studies

Study System	Method Used	Key Performance Outcome	Genetic Markers	Reference
East/Southeast Asian populations	XGBoost	95.6% ancestry classification accuracy	2,000 AISNPs	[20]
East/Southeast Asian populations	Locator (deep neural network)	Geographic localization nearly equivalent to 597,569 SNPs	2,000 AISNPs	[20]
Purple cone spruce (Picea) homoploid hybrid speciation	MSC-I model	Effectively reconstructed hybrid speciation history	Multi-locus sequence data	[54]
Bacterial core genomes (50 genera)	Phylogenomic incongruence	Average 2.76% median introgressed core genes (up to 14% in Escherichia-Shigella)	Core genome	[51]

Experimental Protocols for Introgression Detection

Protocol 1: Tree-Based Introgression Detection with MSC Models

Application Context: This protocol is ideal for testing specific gene flow hypotheses between species with known phylogenetic relationships, particularly when the direction and timing of introgression are of interest.

Detailed Workflow:

• Locus Selection and Sequencing: Select multiple independent loci (UCEs, exome captures, or RADseq) with minimal recombination within loci. The MSC model assumes no recombination within loci but free recombination between them [54].
• Sequence Alignment and Quality Control: Generate multiple sequence alignments for each locus. Implement strict quality filters and assess each locus for phylogenetic signal.
• Model Selection: Choose between MSC-I (discrete introgression) and MSC-M (continuous migration) based on biological knowledge. MSC-I is more parameter-efficient for pulse-like events, while MSC-M better suits ongoing gene flow [54].
• MCMC Sampling and Convergence: Run Bayesian MCMC analyses (e.g., in BPP) with sufficient iterations. Assess convergence using multiple chains and effective sample size diagnostics.
• Model Comparison: Compare marginal likelihoods of different gene flow models (e.g., isolation, migration, introgression) using Bayes factors or similar metrics.
• Sensitivity Analysis: Test robustness to prior distributions and model assumptions, particularly for introgression probability (φ) and migration rate (M) parameters.

Interpretation Guidelines: Strong evidence for introgression typically requires Bayes factors >10 in favor of models with gene flow. However, be aware that mis-assignment of gene flow to incorrect lineages can cause large biases in parameter estimates [54].

Protocol 2: SNP-Based Ancestry Inference with Machine Learning

Application Context: This protocol applies to fine-scale ancestry inference and geographic localization, particularly in forensic science, biogeography, or studies of admixed populations.

Detailed Workflow:

• Reference Panel Curation: Compile genotype data from reference populations with precise geographical or cultural labels. Implement quality control (e.g., call rates >90%, MAF >1%, HWE p-values >0.001) [20].
• AISNP Panel Design: Identify ancestry-informative SNPs using measures like Rosenberg's In statistic. Create nested panels (e.g., 50 to 2,000 SNPs) for efficiency-resolution tradeoffs [20].
• Machine Learning Model Training: Train multiple classifiers (XGBoost, RF, SVM, CNN) using k-fold cross-validation. Optimize hyperparameters through grid search [20].
• Model Evaluation: Assess performance using accuracy, AUC-ROC, and cross-validation error. The optimized XGBoost model has achieved 95.6% accuracy with 2,000 AISNPs [20].
• Geographic Prediction (Optional): For geographic localization, train deep neural networks like Locator on unphased genotypes to predict latitude and longitude directly [20].
• Validation on Admixed Individuals: Apply the trained model to individuals with unknown or admixed ancestry, providing classification probabilities and uncertainty estimates.

Interpretation Guidelines: High accuracy (>95%) is achievable with optimized SNP panels and machine learning. For geographic prediction, performance with reduced AISNP panels can approach that of genome-wide data [20].

Research Reagent Solutions

Table 3: Essential Research Reagents and Computational Tools

Reagent/Tool	Category	Primary Function	Application Context
BPP Software	Tree-based analysis	Bayesian MCMC implementation of MSC-I and MSC-M models	Phylogenomic inference of species divergence with gene flow [54]
AISNP Panels	Genetic markers	Ancestry-informative SNPs selected for maximum population differentiation	Reduced representation genotyping for ancestry inference [20]
XGBoost	Machine learning classifier	Gradient boosting framework for classification tasks	High-accuracy ancestry prediction from SNP data [20]
Locator	Deep neural network	Geographic coordinate prediction from genetic data	Inferring spatial origins from genotypes without phased data [20]
ADMIXTURE	Population genetics	Model-based estimation of ancestry components	Unsupervised clustering of individuals into K ancestral populations [20]
PLINK	Data management	Genotype data quality control and format conversion	Processing and filtering SNP data before analysis [20]

The choice between tree-based and SNP-based introgression tests depends critically on divergence time, data type, and research objectives. Tree-based MSC methods are superior for deep evolutionary questions where estimating parameters like divergence times, direction of gene flow, and introgression probabilities is essential. They perform best with multi-locus sequence data and when the species phylogeny is of primary interest. SNP-based machine learning methods excel in applications requiring fine-scale ancestry decomposition or geographic localization, particularly with closely related populations or admixed groups. They offer practical advantages with reduced marker sets and can achieve remarkably high accuracy with optimized panels.

For comprehensive evolutionary studies, a hierarchical approach may be optimal: using tree-based methods to establish the species framework and major gene flow events, then applying SNP-based methods to resolve fine-scale population structure and adaptive introgression patterns. As both methodologies continue to advance, their complementary strengths provide evolutionary biologists with an increasingly powerful toolkit for deciphering the complex history of gene flow that shapes biodiversity.

In the rapidly evolving field of evolutionary genomics, researchers frequently face a choice between numerous computational methods for detecting introgression. Benchmarking studies provide a rigorous framework for comparing the performance of different methods using well-characterized datasets to determine their strengths and weaknesses, ultimately offering evidence-based recommendations for method selection [55]. The reliability of scientific conclusions in comparative studies of tree-based versus SNP-based introgression tests depends fundamentally on the rigorous validation of analytical pipelines. Without proper benchmarking, methodological artifacts can be easily misinterpreted as biological signals, leading to flawed evolutionary inferences.

Simulation-based benchmarking offers unique advantages for pipeline validation by providing known ground truth against which method performance can be quantitatively assessed. Unlike real empirical datasets where the true evolutionary history is unknown, simulations allow researchers to precisely control parameters such as divergence times, population sizes, selection strengths, and migration rates [56] [57]. This controlled environment enables researchers to systematically evaluate how different methods perform across various evolutionary scenarios that might be encountered in empirical studies. However, the design and implementation of these simulation studies require careful consideration to provide accurate, unbiased, and informative results that genuinely reflect methodological performance under realistic conditions.

This guide synthesizes essential principles for designing, executing, and interpreting benchmarking studies for introgression detection pipelines, with particular emphasis on the comparative performance evaluation of tree-based and SNP-based methods. By following structured benchmarking approaches, researchers can generate reliable evidence to guide method selection and implementation for specific research questions in evolutionary genomics.

Fundamental Principles of Effective Benchmarking

Defining Clear Objectives and Scope

The foundation of any successful benchmarking study is a precisely defined purpose and scope established at the outset. Benchmarking studies generally fall into three broad categories: (1) those conducted by method developers to demonstrate the advantages of a new approach; (2) neutral studies performed by independent groups to systematically compare existing methods; and (3) community-organized challenges that establish standardized evaluations [55]. Each category demands different design considerations, particularly regarding method selection and comprehensiveness.

For neutral benchmarks comparing tree-based and SNP-based introgression methods, the study should strive to be as comprehensive as possible within resource constraints. The research team should maintain approximate familiarity with all included methods to minimize perceived bias and reflect typical usage by independent researchers [55]. Alternatively, involving original method authors can ensure each method is evaluated under optimal conditions, though this approach requires careful management to maintain overall balance in the research team. When authors of particular methods decline to participate, this should be explicitly reported to provide full context for interpretation of results.

Selection of Methods for Comparison

Method selection should be guided by the benchmark's purpose and scope. A comprehensive neutral benchmark should include all available methods for a specific type of introgression analysis, functioning as a systematic review of the field. Practical constraints often necessitate defining explicit inclusion criteria, such as methods with freely available software implementations, compatibility with common operating systems, and installability without excessive troubleshooting [55]. These criteria must be applied uniformly without favoring specific methods, and exclusion of widely used tools should be scientifically justified.

For benchmarks focused on new method development, it is generally sufficient to compare against a representative subset of existing methods, including current best-performing approaches, simple baseline methods, and any widely used standards in the field [55]. The selection should enable accurate and unbiased assessment of the new method's relative merits compared to the current state-of-the-art. In fast-moving fields, benchmarks should be designed to allow extensions as new methods emerge, ensuring ongoing relevance.

Table 1: Method Selection Criteria for Benchmarking Studies

Criterion	Comprehensive Benchmark	Method Development Benchmark
Scope	All available methods	Representative subset
Inclusion Basis	Literature review	State-of-the-art and baseline methods
Practical Requirements	Freely available, installable software	Comparable computational requirements
Documentation	Summary table of all methods	Focus on differences from existing methods
Extensibility	Framework for future additions	Planned updates as new methods emerge

Dataset Selection and Simulation Design

The choice of reference datasets constitutes one of the most critical design decisions in benchmarking. Simulated data offer the significant advantage of known ground truth, enabling precise quantification of performance metrics [55]. However, simulations must accurately reflect relevant properties of real biological data to provide meaningful insights. Empirical summaries of both simulated and real datasets should be compared to verify simulation realism, with specific metrics chosen based on context—for example, dropout profiles and dispersion-mean relationships for RNA-seq data, or site frequency spectra and linkage disequilibrium patterns for population genomic data [55].

A well-designed benchmark incorporates a variety of datasets representing different evolutionary scenarios and demographic histories. For example, a recent benchmark of adaptive introgression methods simulated scenarios inspired by diverse biological systems including humans, Iberian wall lizards (Podarcis), and bears (Ursus), varying parameters such as divergence time, selection strength, timing of gene flow, effective population size, and recombination rates [57]. This approach enables researchers to assess whether method performance generalizes across different evolutionary contexts or is optimized for specific biological systems.

Benchmarking Tree-Based vs. SNP-Based Introgression Methods

Methodological Foundations and Key Differences

Tree-based and SNP-based methods for detecting introgression differ fundamentally in their underlying approaches and data requirements. Tree-based methods typically operate within the framework of the multispecies coalescent (MSC) model, using data from one sample per species to infer gene tree frequencies and branch lengths [56]. These methods leverage the fact that gene tree heterogeneity—variation in tree topologies across genomic loci—can result from both incomplete lineage sorting (ILS) and introgression. The minimal data requirement for powerful tests of introgression based on gene tree discordance is a rooted triplet of species (or an unrooted quartet) [56].

SNP-based methods generally operate on biallelic site patterns or allele frequency differences between populations. These include summary statistics, likelihood methods, and machine learning approaches that identify genomic regions with unusual patterns of variation suggestive of introgression. Methods like Q95, VolcanoFinder, MaLAdapt, and Genomatnn represent different statistical approaches to detecting adaptive introgression, each with particular strengths and limitations [57].

Performance Comparison Across Evolutionary Scenarios

Recent benchmarking efforts have revealed that method performance varies significantly across different evolutionary scenarios. In a comprehensive evaluation of adaptive introgression detection methods, the relatively simple Q95 statistic performed remarkably well across most scenarios, often outperforming more complex machine learning approaches, especially when applied to species or demographic histories different from those used in training the models [57]. This finding highlights the tension between methodological sophistication and generalizability, particularly for machine learning approaches that may overfit to their training data.

The performance of different methods depends critically on factors such as divergence time, selection strength, timing of gene flow, effective population size, and recombination landscape [57]. For example, methods developed and trained specifically on human genomic data (particularly admixture between Homo sapiens, Neanderthals, and Denisovans) may perform poorly when applied to other biological systems with different demographic histories. This underscores the importance of tailoring detection approaches to the evolutionary history of the study system rather than relying on universal "best" methods.

Table 2: Performance Characteristics of Introgression Detection Methods

Method Type	Key Strengths	Key Limitations	Optimal Use Cases
Tree-Based Methods	Robust to selection; well-characterized statistical properties; intuitive biological interpretation	Require accurate gene tree estimation; computationally intensive for large datasets; sensitive to model misspecification	Deep phylogenetic scales; situations with substantial ILS; when species tree is well-established
SNP-Based Summary Statistics	Computational efficiency; simple implementation and interpretation; minimal assumptions about demographic history	Limited power for complex demography; may not distinguish different sources of signal; limited characterization capability	Initial screening; large genomic datasets; rapid exploratory analysis
Machine Learning Approaches	Potential to capture complex patterns; integration of multiple signals; high performance in trained scenarios	Risk of overfitting; limited interpretability; performance depends on training data similarity	Well-characterized systems with sufficient training data; integration of multiple genomic features

Experimental Design for Method Comparison

A robust experimental design for comparing tree-based and SNP-based introgression methods should incorporate several key elements. First, it should include a range of simulation scenarios covering different demographic histories, selection regimes, and genomic architectures. These scenarios should reflect realistic evolutionary contexts rather than only idealized conditions. Second, the design should explicitly test method performance under model violations to assess robustness. Third, it should evaluate methods across a spectrum of data quality conditions, including varying sequence lengths, missing data patterns, and sequencing error rates.

A particularly valuable approach is to use simulations that mirror the evolutionary histories of specific empirical systems for which reliable independent evidence of introgression exists. This enables not only comparison of method performance under controlled conditions but also validation of conclusions against biological reality. For example, simulations parameterized using estimates from well-studied systems such as Helianthus sunflowers, Mus mice, or Picea spruce trees provide evolutionary realistic contexts for method evaluation [58] [59].

Implementation Framework for Benchmarking Studies

Workflow Design and Execution

The following diagram illustrates a comprehensive benchmarking workflow that integrates key principles of effective pipeline validation:

Performance Metrics and Evaluation Criteria

Selecting appropriate evaluation criteria is essential for meaningful method comparison. Performance metrics should capture different aspects of method performance, including overall accuracy, power to detect true introgression, false positive rates, and precision in characterizing introgression parameters. For classification-based methods (e.g., detecting genomic regions affected by introgression), standard metrics include sensitivity, specificity, precision, and area under the receiver operating characteristic curve (AUC-ROC) [57].

For methods that estimate continuous parameters (e.g., introgression proportion, timing, or selection coefficients), evaluation should include measures of estimation accuracy such as bias, mean squared error, and calibration. Additional practical considerations include computational efficiency, memory requirements, scalability to large genomic datasets, and usability factors such as documentation quality and ease of implementation [55]. No single metric captures all relevant aspects of performance, so a multifaceted evaluation approach is necessary.

Table 3: Essential Performance Metrics for Introgression Detection Benchmarks

Metric Category	Specific Metrics	Interpretation
Classification Performance	Sensitivity, Specificity, Precision, F1-score, AUC-ROC	Overall discrimination ability between introgressed and non-introgressed regions
Parameter Estimation	Bias, Mean Squared Error, Coverage Probability, Calibration	Accuracy and reliability of parameter estimates (proportion, timing, strength)
Robustness	Performance under model misspecification, missing data, sequencing error	Reliability under non-ideal conditions commonly encountered in empirical studies
Computational Efficiency	Runtime, Memory usage, Scalability with dataset size	Practical feasibility for typical research applications
Usability	Installation success, Documentation quality, Error handling	Ease of implementation and use by researchers with varying expertise

Successful benchmarking requires careful selection of computational tools and resources. The following table outlines key components of an effective benchmarking toolkit:

Table 4: Essential Research Reagents for Benchmarking Introgression Detection Methods

Tool Category	Specific Tools/Frameworks	Primary Function
Simulation Software	msprime, SLiM, stdpopsim	Generate synthetic genomic data with known evolutionary histories
Method Implementation	Specific tree-based (e.g., HyDe, D-statistics) and SNP-based (e.g., Q95, VolcanoFinder) tools	Execute introgression detection methods on simulated and empirical data
Performance Evaluation	scikit-learn, custom evaluation scripts	Calculate performance metrics and generate comparative visualizations
Workflow Management	Snakemake, Nextflow	Automate and reproduce complex benchmarking pipelines
Visualization	matplotlib, seaborn, ggplot2	Create publication-quality figures summarizing benchmarking results

Advanced Considerations in Benchmark Design

Addressing Method-Specific Biases and Limitations

Benchmarking studies must carefully address potential biases that can distort performance comparisons. A common pitfall is uneven parameter tuning across methods—extensively optimizing parameters for one method while using default settings for others [55]. To ensure fair comparisons, all methods should be given comparable opportunities for optimization, either through automated parameter searches or by involving method developers who can provide optimal settings for specific scenarios.

Another significant challenge is that methods may not be directly comparable if they were designed for different tasks or make different assumptions. For example, some methods assume a specific phylogenetic history, while others are designed for population-level data with continuous gene flow. Still others focus specifically on detecting adaptive introgression rather than neutral introgression [57]. The benchmarking design should clearly acknowledge these differences and, when appropriate, include sub-analyses that group methods by their intended applications and theoretical foundations.

Reproducibility and Extensibility

Ensuring that benchmarking studies are reproducible and extensible is essential for their long-term value. Best practices include providing complete code and documentation, containerization of software environments using Docker or Singularity, and depositing both scripts and results in persistent repositories with digital object identifiers [57]. These practices enable other researchers to verify findings and extend the benchmark as new methods emerge.

A particularly valuable approach is to design benchmarking frameworks that can easily incorporate additional methods, datasets, or performance metrics. This might involve standardized input/output formats, modular code architecture, and clear documentation for contributors. Community challenges, such as those organized by the DREAM consortium, provide excellent models for this approach, though they require substantial organizational investment [55].

Interpretation and Reporting of Results

Contextualizing Performance Differences

When interpreting benchmarking results, it is essential to consider that performance differences between methods may be minor or context-dependent. Rather than declaring a single "best" method, a more nuanced approach is to identify a set of high-performing methods and highlight their different strengths and tradeoffs [55]. This might include differences in sensitivity to particular evolutionary scenarios, computational requirements, or usability factors that make methods more or less suitable for specific research contexts.

Performance should be interpreted in light of the benchmark's limitations, including the specific scenarios tested, the metrics emphasized, and any methods that were excluded or encountered technical difficulties. Transparent reporting of these limitations helps users understand the generalizability of the findings and potential areas where method performance might differ from what was observed in the benchmark.

Guidelines for Method Selection

The ultimate goal of benchmarking studies is to provide practical guidance for researchers selecting methods for their specific applications. Effective guidelines should consider multiple factors beyond raw performance, including:

• Biological context: Does the method make assumptions appropriate for the study system?
• Data requirements: Are the input data requirements feasible for the available data?
• Computational resources: Is the method computationally practical given available resources?
• Interpretability: Are the outputs biologically interpretable and suitable for addressing the research question?
• Usability: Is the method well-documented and accessible to researchers with appropriate expertise?

For example, a researcher working with non-model organisms with limited genomic resources might prioritize different methods than a researcher working with well-characterized model systems with high-quality reference genomes and extensive population sampling.

Rigorous benchmarking using simulations provides an essential foundation for validating introgression detection pipelines and comparing the performance of tree-based and SNP-based methods. By following structured approaches to benchmark design, implementation, and interpretation, researchers can generate reliable evidence to guide method selection and application. The rapidly evolving nature of genomic methods necessitates that benchmarking become an ongoing community effort rather than a one-time assessment, with frameworks designed for extensibility as new methods and evolutionary questions emerge.

The comparative performance of tree-based and SNP-based methods depends critically on evolutionary context, with no single approach dominating across all scenarios. This context-dependence underscores the importance of tailoring method selection to specific research questions and biological systems rather than relying on universal performance rankings. As the field advances, continued development and refinement of benchmarking standards will play a crucial role in ensuring the reliability and reproducibility of evolutionary genomic inferences.

Benchmarking Performance: Empirical and Simulation-Based Evidence

In genetic epidemiology and phylogenetics, accurately identifying true positive signals (statistical power) while controlling for false positives (Type I error rates) is a fundamental challenge. Simulation studies provide a controlled environment to evaluate the performance of various statistical methods, guiding researchers to select the most appropriate test for their specific data and hypotheses. This guide objectively compares the performance of tree-based and SNP-based methods, two dominant approaches in genetic analysis, focusing on their application in detecting introgression and genetic associations. We summarize empirical evidence from recent simulation studies, provide detailed experimental protocols, and visualize key workflows to inform researchers and drug development professionals.

Comparative Performance of Statistical Methods

The table below synthesizes key findings from recent simulation studies, directly comparing the statistical power and Type I error rates of tree-based and SNP-based methods across various genetic analyses.

Table 1: Performance Comparison of Tree-based and SNP-based Methods from Simulation Studies

Analysis Type	Method Category	Specific Method	Statistical Power	Type I Error Rate	Key Simulation Finding
Genetic Risk Score (GRS) Construction [60]	Tree-based	Random Forests, Logic Bagging	Higher (especially with epistasis)	Comparable to or controlled vs. linear models	Outperformed elastic net in most scenarios, particularly with epistatic interactions.
	Regularized Regression	Elastic Net	Lower	Comparable to or controlled vs. tree-based	Lead to inferior results in most cases, even with only marginal effects.
Distant Relationship Inference [61]	Likelihood-based	Likelihood Ratio (LR)	Highest (with <20k SNPs)	Controlled	Most powerful method with sparse SNP data; easily adapted for non-pairwise tests.
	Segment/Kinship-based	Windowed Kinships, Segment Approach	High (with >20k SNPs)	Controlled	Equally powerful as LR only when very dense SNP data (>20k markers) are available.
	Method-of-Moments	Kinship Coefficient Estimators	Moderate (for <4th degree)	Performance declines beyond 4th degree	Performs well for lower-degree relationships but less so for distant relatives.
Species Tree & Parameter Inference [62]	Coalescent-based (with error)	BPP (with genotyping errors)	Reduced (high error, low depth)	Biased (high error, low depth)	High error rates (e=0.01) and low depth (<10x) reduce power and bias parameter estimates.
	Coalescent-based (ideal)	BPP (no errors)	High (baseline)	Controlled (baseline)	At low error rate (e=0.001, Phred 30), inference is little affected even at ~3x depth.

Performance Analysis and Key Trends

• Context-Dependent Superiority: No single method is universally superior. Tree-based methods like random forests and logic bagging excel in detecting complex, non-linear genetic interactions (epistasis) for traits such as disease risk [60]. In contrast, SNP-based likelihood methods demonstrate higher power for inferring relationships like relatedness or phylogeny, especially with limited genetic markers [61].

• Impact of Data Quality: The performance of sophisticated methods is highly dependent on data quality. In phylogenetic inference, genotyping errors at a rate of e=0.01 combined with low sequencing depth (<10x) can significantly reduce the power of species tree estimation and introduce substantial bias in population parameters [62].

• Sample Size and Marker Density Trade-offs: For relationship inference, the likelihood ratio (LR) method is most powerful with smaller SNP panels (<20,000 markers), while segment-based approaches require very dense genomic data (>20,000 markers) to achieve comparable power [61].

Detailed Experimental Protocols

Protocol 1: Simulating GRS Performance with Tree-Based Methods

This protocol is derived from studies evaluating genetic risk score (GRS) construction for binary traits [60].

Data Generation and Simulation Setup

• Genotype Simulation: Simulate a population of N individuals with p biallelic Single Nucleotide Polymorphisms (SNPs). Each SNP is coded as 0 (homozygous reference), 1 (heterozygous), or 2 (homozygous variant).
• Phenotype Simulation: Generate a binary trait (case/control) influenced by the SNPs. The underlying model can include:
  • Marginal effects: Additive contributions from individual SNPs.
  • Epistatic effects: Non-linear interaction effects between multiple SNPs.
  • Heritability and Prevalence: Set the overall heritability of the trait and the disease prevalence in the simulated population.
• Data Partitioning: Split the simulated dataset into training (≈50%) and testing (≈50%) sets to evaluate model generalizability [60].

Model Fitting and Comparison

• Implement Methods:
  • Tree-based: Train a Random Forest model. Optionally, employ a modified version like "logic bagging" to improve stability.
  • SNP-based (Benchmark): Train a regularized regression model such as Elastic Net.
• Hyperparameter Tuning: Conduct extensive joint hyperparameter optimization for all methods using the training set. For Random Forests, this includes the number of trees, tree depth, and number of features considered per split. For Elastic Net, tune the L1/L2 regularization mixing parameter.
• Performance Evaluation: Apply the fitted models to the test set. Calculate the Area Under the Curve (AUC) to measure predictive power and perform statistical tests to ensure Type I error rates are controlled at the desired level (e.g., α=0.05).

Protocol 2: Evaluating Type I Error in Phylogenetics under Genotyping Errors

This protocol assesses the impact of genotyping and sequencing errors on species tree inference, a key application of SNP-based coalescent methods [62].

Simulating True Genomic Sequences

• Coalescent Simulation: Use the multispecies coalescent (MSC) model to simulate the true species phylogeny and population genetic parameters (divergence times, population sizes, migration rates).
• Generate True Sequences: For each simulated genealogical history, evolve DNA sequences along the gene trees to create true, error-free multi-locus aligned sequences for all samples.

Introducing Genotyping Errors

• Model Read Depth: Simulate realistic read depths for each site in the sequence alignment using a Markov model. The depth for a site is drawn from a Beta distribution, with adjacent sites having correlated depths (e.g., autocorrelation parameter p = 0.9) to mimic real sequencing data [62].
• Introduce Base-Calling Errors: For each site, given the true genotype and the simulated read depth, simulate sequencing reads by multinomial sampling. Incorporate base-calling errors at a specified rate (e.g., e=0.001 for high-quality data vs. e=0.01 for low-quality data).
• Call Genotypes: From the simulated reads, call genotypes using a maximum likelihood (ML) estimator. The output is a dataset of unphased diploid sequences containing genotyping errors.

Inference and Error Measurement

• Species Tree and Parameter Estimation: Analyze the error-containing dataset using Bayesian coalescent-based software like Bpp, ignoring the presence of genotyping errors in the model.
• Calculate Performance Metrics:
  • Power: Measure the proportion of simulations where the true species tree topology is correctly recovered.
  • Bias: Calculate the difference between the estimated population parameters (e.g., population size θ, divergence time τ) and their known true values from the simulation.
  • Type I Error Rate: When testing for a specific event (e.g., gene flow), simulate data without such an event and calculate the proportion of simulations where the test falsely rejects the null hypothesis.

Visualizing Simulation Workflows

Workflow for Genetic Risk Score Comparison

The diagram below outlines the logical flow and comparison points for Protocol 1, simulating the performance of tree-based and SNP-based GRS methods.

Diagram 1: GRS Method Comparison Workflow

Workflow for Phylogenetic Error Impact

The diagram below illustrates the process of evaluating the impact of genotyping errors on SNP-based coalescent analysis, as described in Protocol 2.

Diagram 2: Phylogenetic Error Impact Workflow

The Scientist's Toolkit

Table 2: Essential Research Reagents and Computational Tools

Item/Reagent	Function/Purpose	Example/Note
Ancestry-Informative SNP (AISNP) Panels	Sets of pre-selected SNPs with high population differentiation, used for ancestry inference and localization.	Nested panels (50-2,000 SNPs) can be combined with machine learning (e.g., XGBoost) for high-accuracy inference [20].
Global Screening Array (GSA)	A commercial SNP array commonly used in direct-to-consumer genetics and large-scale screening studies.	Serves as a base panel; can be expanded with specialized forensic or kinship markers (e.g., FORCE, Kintelligence) [61].
PLINK Software	A whole-genome association analysis toolset used for extensive data quality control and management.	Used for pruning SNPs in linkage disequilibrium (LD) and basic genotype data processing [61] [20].
ADMIXTURE Software	A tool for estimating ancestry components and population structure from genotype data.	Used to determine individual ancestry proportions, often as input for supervised classification or to define genetic groups [20].
Bpp Software	A Bayesian program for phylogenetic inference and population parameter estimation under the multispecies coalescent.	Used to infer species trees, divergence times, and gene flow; sensitive to genotyping errors at low sequencing depths [62].
ped-sim Software	A script for simulating pedigree genetic data with realistic recombination and crossover interference.	Used to generate genotype data for pairs of relatives (e.g., siblings, cousins) for kinship analysis [61].
Locator Model	A deep neural network framework that predicts geographic coordinates (latitude/longitude) directly from genetic data.	Can achieve high precision with a small number of AISNPs (~2,000), nearly matching genome-wide data performance [20].

The detection of introgressed genomic regions—where genetic material has been transferred between species or populations—is fundamental to understanding evolutionary processes such as adaptation and speciation. The choice of analytical method can significantly impact the accuracy and biological relevance of these findings. This guide provides an objective comparison between tree-based phylogenetic methods and SNP-based summary statistic methods for detecting introgression. Evidence from simulated and empirical studies across diverse taxa indicates that while SNP-based methods offer computational efficiency, tree-based approaches generally provide superior robustness, particularly when dealing with deep evolutionary divergences, low levels of introgression, and the challenge of distinguishing introgression from incomplete lineage sorting.

Introgression, the transfer of genetic material between species through hybridization and backcrossing, is a recognized force in evolution, with implications for adaptation and the emergence of novel traits [38]. Accurately identifying introgressed regions in genomic data is crucial for testing hypotheses about evolutionary history and selection. The two primary classes of methods for this task are SNP-based summary statistics and tree-based phylogenetic methods.

• SNP-based Summary Statistics: These methods, including dXY, dmin, Gmin, and RNDmin, reduce genetic data to single numerical values that measure divergence or similarity between populations. Regions with significantly elevated similarity (e.g., low dXY or RNDmin) are flagged as potential introgression candidates [38]. Their strengths are computational speed and simplicity.
• Tree-Based Phylogenetic Methods: These methods leverage the full evolutionary history of sequences, typically represented by a phylogenetic tree. They identify introgression by detecting topological inconsistencies between the evolutionary trees of different genomic regions (gene trees) and the expected species tree [51]. While computationally intensive, they more fully use implicit phylogenetic information.

The following sections compare these approaches experimentally, highlighting their performance in detection power, robustness, and applicability to ancient systems.

Performance Comparison: Experimental Data

A direct comparison of a tree-based method (Likelihood Score Statistic - LSS) and a non-tree-based method (pooled t-test) on simulated systolic blood pressure data across five genes revealed critical performance differences [6].

Table 1: Performance Comparison of Tree-Based vs. Non-Tree-Based Methods for QTM Detection

Gene (Effect Size)	Method	Type I Error Rate	Detection Performance
TNN (Large)	Tree-Based (LSS)	0.010	Well
	Non-Tree-Based (t-test)	>0.05	Well
LEPR (Large)	Tree-Based (LSS)	0.045	Well
	Non-Tree-Based (t-test)	>0.05	Well
GSN (Small)	Tree-Based (LSS)	0.020	Low Power
	Non-Tree-Based (t-test)	>0.05	Low Power

The data show that while both methods successfully detected genes with large effect sizes, the tree-based LSS method maintained a significantly lower Type I error rate (i.e., fewer false positives) across all genes compared to the t-test [6]. This demonstrates the superior statistical robustness of the tree-based approach in controlling for false discoveries. For genes with weaker signals, both methods showed low power, indicating a universal challenge in detecting subtle introgression events.

Methodologies and Workflows

Key SNP-Based Statistics

SNP-based methods are often designed to be robust to certain confounding factors. The following statistics are commonly used [38]:

• dXY: The average number of sequence differences between all sequences in two taxa. Low values may indicate introgression but can be confounded by regions of low mutation rate.
• dmin: The minimum sequence distance between any pair of haplotypes from two taxa. It is sensitive to rare, recent introgression events but is also susceptible to false positives from mutation rate variation.
• RNDmin: A normalization of dmin to account for variation in the neutral mutation rate. It is calculated as the quotient of dmin and the average distance from each taxon to an outgroup (dout). This makes it more robust than dmin alone.
• Gmin: Defined as dmin/dXY, this statistic also normalizes for variable evolutionary rates among loci and is sensitive to recent migration.

Tree-Based Method Workflow

The tree-based Likelihood Score Statistic (LSS) approach follows a detailed phylogenetic workflow [6]:

• Tree Estimation: At each SNP, a phylogenetic tree (Θ) is estimated from the haplotype data. Computational efficiency is often maintained by using a broad-scale estimate of the tree, considering it as a set of k clusters.
• Model Assumption: The quantitative trait values are assumed to follow a multivariate normal distribution. The mean structure is defined by the mean trait value for each cluster (μ₁, μ₂, ..., μₖ). The covariance structure (V(Θ)) is defined by the shared evolutionary history in the estimated tree, with the covariance between two observations proportional to the length of their shared lineages.
• Scoring: The LSS is calculated as a penalized likelihood score, specifically the maximum of 2 ln L(μ̂, σ̂² | y, V(Θ), Θ) - k ln n, taken over the number of clusters k. This score effectively compares the fit of the phylogenetic model to the data while penalizing for model complexity.

Robustness and Performance in Ancient Systems

The robustness of tree-based methods becomes particularly evident in evolutionarily deep or "ancient" systems, where signals of introgression are faint and confounded by other processes.

• Power to Detect Rare and Ancient Introgression: A simulation study introducing the tree-informed RNDmin statistic found it offered a modest increase in power over other related SNP-based tests (dmin, Gmin). All tests, however, had high power only when migration was recent and strong [38]. This suggests that for older, weaker introgression events—common in ancient systems—tree-based methods hold an advantage.
• Distinguishing Introgression from Incomplete Lineage Sorting (ILS): A major challenge in deep phylogenies is distinguishing introgression from ILS, where gene tree discordance arises from the random retention of ancestral polymorphisms. SNP-based statistics like dmin can be misled by ILS, whereas model-based phylogenetic methods explicitly model the coalescent process to separate these confounding signals.
• Application to Bacterial Species Borders: Despite the prevalence of gene flow (homologous recombination) in bacteria, a 2025 study found that core genome phylogenetic trees clearly delineate most species. While introgression was detected in core genes (averaging ~2-8% across 50 lineages), it rarely created "fuzzy" species borders [51]. This demonstrates the power of tree-based phylogenomics to identify stable evolutionary units despite ongoing genetic exchange.

Table 2: Suitability of Methods for Different Evolutionary Contexts

Evolutionary Context	Recommended Method	Rationale
Recent, Strong Introgression	SNP-based (e.g., dmin, RNDmin)	High power and computational efficiency for clear signals.
Deep Divergence / Ancient Introgression	Tree-based (e.g., LSS, Phylogenomics)	Superior at handling ILS and detecting weaker, older signals.
Closely Related Species with Porous Borders	Tree-based (BSC-species definition)	Effectively identifies cohesive genetic clusters despite gene flow [51].
Analysis Requiring High Throughput	SNP-based	Faster computation for genome-wide scans; robustness can be added via normalization (e.g., RNDmin).

The Scientist's Toolkit: Essential Research Reagents and Software

Successful introgression analysis relies on a suite of bioinformatics tools and reference data.

Table 3: Key Research Reagents and Software for Introgression Analysis

Tool / Resource	Type	Primary Function	Application Note
Beagle [63]	Software	Genotype imputation and phasing.	Critical for handling missing data in low-quality samples (e.g., degraded remains). Uses HMMs for prediction.
PAML [64]	Software Package	Phylogenetic analysis by maximum likelihood.	Used for ancestral sequence reconstruction and likelihood calculations under evolutionary models.
Lazarus [64]	Software Package	Topological empirical Bayesian analysis.	Integrates ancestral state reconstructions over a distribution of possible trees to incorporate phylogenetic uncertainty.
1000 Genomes Project [63]	Reference Dataset	Catalog of human genetic variation.	Serves as a crucial haplotype reference panel for imputation and population genetic analysis.
ForenSeq Kintelligence Kit [65]	Commercial Panel	Targeted SNP amplification (10,230 SNPs).	Enables kinship and bioancestry analysis from degraded DNA where STR methods fail.
ANI (Average Nucleotide Identity) [51]	Bioinformatic Metric	Quantifies genome-wide sequence similarity.	A ≥95% ANI threshold is often used to empirically circumscribe bacterial species.

The choice between tree-based and SNP-based methods for introgression detection is context-dependent. SNP-based summary statistics like RNDmin and Gmin are powerful, fast, and sufficiently robust for identifying recent and strong introgression signals. However, for studies focused on ancient systems, deep divergences, or situations where distinguishing introgression from ILS is critical, tree-based phylogenetic methods demonstrate superior robustness and statistical reliability. Their ability to explicitly model evolutionary history and control false positives makes them the more rigorous choice for probing complex evolutionary histories, despite their greater computational demands. As genomic datasets grow in size and complexity, the nuanced application of both classes of methods will continue to be essential for unraveling the history of life.

In the field of population genomics, accurately identifying introgressed genetic material involves two distinct but connected goals: the detection of a statistical signal indicating introgression has occurred, and the precise localization of the specific causal loci responsible for adaptation. The performance of methods in these tasks varies significantly, with approaches generally divided into tree-based methods, which use phylogenetic relationships, and SNP-based (or allele-frequency) methods, which analyze patterns of shared alleles. This guide provides a structured comparison of these methodologies, detailing their experimental protocols, performance under different evolutionary scenarios, and the specific reagents required for their implementation, to inform researchers in selecting the optimal tool for their investigations.

Introgression, the transfer of genetic material between species through hybridization and backcrossing, is increasingly recognized as a fundamental evolutionary force [3] [42]. It can provide a reservoir of genetic variation that facilitates rapid adaptation to new environments, potentially faster than through de novo mutation alone [3]. The analytical process for studying introgression typically involves two phases, which are crucial to distinguish:

• Detection refers to the initial identification that a historical introgression event has occurred between two taxa. It answers the question, "Did these species hybridize?"
• Localization is the subsequent, finer-scale process of pinpointing the exact genomic regions or individual loci that were introgressed and, in cases of adaptive introgression, identifying those under selection [7].

The choice of method is particularly critical when studying ancient introgression, which occurred in the distant past, as the performance and reliability of different tools can vary dramatically in these contexts [42]. The following sections objectively compare the two predominant methodological frameworks.

Comparative Analysis of Methodological Approaches

The two primary classes of methods for introgression analysis rely on different types of genomic data and underlying assumptions. The table below summarizes their core characteristics.

Table 1: Core Characteristics of Tree-based and SNP-based Introgression Methods

Feature	Tree-based Methods	SNP-based Methods
Primary Data	Genome-wide sets of local phylogenetic trees (gene trees) [11].	Patterns of ancestral (A) and derived (B) alleles across single nucleotide polymorphisms (SNPs) [42].
Key Assumptions	Sequence evolution models are used for tree inference; constant rates are not always assumed [11].	Absence of homoplasy (recurrent mutation) and constant evolutionary rates across lineages [42].
Typical Output	Frequencies of alternative tree topologies; support for a phylogenetic network [11] [42].	Statistics quantifying imbalance in allele sharing (e.g., D-statistic) [42].
Computational Intensity	High, due to the need to infer many phylogenetic trees [11].	Generally lower, as calculations are based on site patterns.

Key SNP-Based Methods and Protocols

ABBA-BABA Test (D-statistic) This is a widely used SNP-based method for detecting introgression [11] [42].

• Experimental Protocol: The analysis requires genomic data from four taxa: two sister populations (P1 and P2), a third population (P3) that is the potential source of introgression, and an outgroup (O) to determine the ancestral allele state. The protocol involves: 1) Whole-genome sequencing or dense genotyping of individuals from the four populations. 2) Variant calling and identification of bi-allelic sites. 3) For each SNP, classifying the pattern in P1, P2, P3, and O as "ABBA" (where P2 and P3 share a derived allele not found in P1) or "BABA" (where P1 and P3 share a derived allele not found in P2). 4) Calculating the D-statistic across all genomic sites: D = (∑ABBA - ∑BABA) / (∑ABBA + ∑BABA). A significant deviation from zero (assessed via block jackknifing) indicates introgression between P3 and either P1 (if D is negative) or P2 (if D is positive) [42].
• Performance Considerations: The D-statistic is powerful for detection but provides limited localization capabilities. Its reliability can be compromised by factors such as ancestral population structure and, critically, variation in evolutionary rates among lineages, which can lead to false-positive signals of introgression in divergent taxa [42].

Adaptive Introgression Classification Methods (e.g., VolcanoFinder, Genomatnn, MaLAdapt) These are more advanced tools designed to localize adaptively introgressed loci.

• Experimental Protocol: These methods typically use a supervised learning framework, training classifiers on simulated genomic data to distinguish between neutral and adaptive introgression signals [7]. A key experimental step is the careful selection of non-adaptive introgressed (non-AI) windows for training, which should include not only neutral regions from unlinked chromosomes but also windows adjacent to the candidate AI window. This accounts for the hitchhiking effect of a selective sweep, which reduces diversity in flanking regions and is crucial for accurate localization [7].
• Performance Considerations: A 2025 performance evaluation found that no single method outperforms all others across every tested evolutionary scenario. Methods based on the Q95(w, y) summary statistic were noted for their efficiency in exploratory studies, while machine learning-based classifiers like Genomatnn can offer high accuracy when properly trained [7].

Key Tree-Based Methods and Protocols

Tree-based D-statistic (D_tree) This is a phylogenetic analogue of the SNP-based D-statistic.

• Experimental Protocol: The workflow involves: 1) Extracting thousands of sequence alignment blocks from a whole-genome alignment, filtering them for completeness and a minimal number of recombination breakpoints [11]. 2) Inferring a maximum likelihood phylogenetic tree for each suitable alignment block using software like IQ-TREE [11]. 3) For a given trio of species (P1, P2, P3), counting the frequencies of the three possible rooted tree topologies among the genome-wide set of gene trees. 4) Calculating D_tree = (Freq_Topology2 - Freq_Topology3) / (Freq_Topology2 + Freq_Topology3), where Topology1 is the inferred species tree. A significant asymmetry indicates introgression [42].
• Performance Considerations: This method is generally more robust to homoplasies (recurrent mutations) than the SNP-based D-statistic, making it potentially more reliable for detection in older systems where rate variation is a concern [11] [42]. However, like its SNP-based counterpart, its power for fine-scale localization is limited.

Species Tree and Network Inference (e.g., ASTRAL, PhyloNet) These methods use genome-wide gene trees to infer broader evolutionary history, including introgression.

• Experimental Protocol: After generating a set of gene trees (as for D_tree), researchers use a tool like ASTRAL to estimate the primary species tree from the distribution of gene tree topologies [11]. Subsequently, a tool like PhyloNet can be used to infer a phylogenetic network that explicitly models introgression events, assessing support for alternative models of diversification with and without hybridization [11].
• Performance Considerations: These approaches are powerful for detecting the presence and direction of major introgression events. Localization of specific introgressed loci can be achieved by identifying genomic regions whose gene tree topologies are inconsistent with the inferred species network.

The logical workflow for selecting and applying these methods is summarized in the following diagram:

Performance Data: Detection vs. Localization

The performance of these methods can be evaluated based on their statistical power for detection and their accuracy for localization, often assessed through simulation studies.

Table 2: Comparative Performance of Introgression Methods

Method (Category)	Detection Power	Localization Accuracy	Key Strengths	Key Limitations
D-statistic (SNP-based)	High for recent introgression [42].	Low; identifies general signal but not specific loci [42].	Fast; easy to implement; works on unphased data.	Prone to false positives from rate variation [42].
Dtree (Tree-based)	Robust for older systems with rate variation [11] [42].	Low; identifies topological asymmetry across genome [11].	Robust to homoplasy; uses full sequence information.	Computationally intensive; requires high-quality alignments [11].
Genomatnn (SNP-based)	High (as a classifier) [7].	High; designed to pinpoint loci [7].	Can capture complex, non-linear patterns.	Requires extensive training data/simulations [7].
Q95(w, y) (SNP-based)	Moderate to High [7].	Moderate; efficient for scans [7].	Efficient for exploratory studies; less computationally demanding.	Performance varies with evolutionary scenario [7].

Impact of Evolutionary Context on Performance

The relative performance of these methods is highly dependent on the biological context.

• Divergence Time and Rate Variation: For deeply divergent taxa (e.g., millions of years), the assumption of constant evolutionary rates is often violated. SNP-based methods like the D-statistic are highly susceptible to false-positive detection signals in these scenarios due to homoplasy [42]. In contrast, tree-based methods are more robust under these conditions, as they are less sensitive to homoplasy and do not implicitly assume rate constancy [11] [42].
• Strength of Selection: The localization of adaptively introgressed loci is more accurate when the selection coefficient is strong, as this produces a clearer genomic signature (e.g., a more pronounced selective sweep) [7].

The Scientist's Toolkit: Essential Research Reagents and Software

Implementing the protocols described above requires a suite of specialized software tools and data resources.

Table 3: Essential Reagents and Software for Introgression Analysis

Item Name	Category	Primary Function	Key Features
Whole-Genome Alignment	Data	A reference-based or reference-free alignment of multiple genomes, serving as the raw data for tree-based methods.	Provides the sequence alignment blocks for phylogenetic inference [11].
IQ-TREE	Software	Infers maximum likelihood phylogenetic trees from sequence alignments.	Modern, rapid tool with model selection; used to generate gene trees [11].
PAUP*	Software	A general-utility program for phylogenetic inference.	Command-line version is used for various phylogenetic analyses [11].
ASTRAL	Software	Estimates the species tree from a set of input gene trees.	Efficient and accurate; accounts for incomplete lineage sorting [11].
PhyloNet	Software	Infers species networks from gene trees in a maximum-likelihood or Bayesian framework.	Models reticulate evolutionary events like introgression and hybridization [11].
Ancestry-informative SNP Panels	Data	A reduced set of SNPs with large frequency differences between populations.	Enables efficient ancestry inference; can be used with machine learning for localization [20].
Genomatnn	Software	A machine learning tool for detecting introgressed loci.	Uses a convolutional neural network; requires training on simulated data [7].
VolcanoFinder	Software	A tool for detecting adaptive introgression.	Based on the site frequency spectrum [7].

The choice between tree-based and SNP-based methods for introgression analysis hinges on the specific research question, particularly the distinction between detection and localization. For the initial detection of introgression, especially among divergent taxa where evolutionary rates may vary, tree-based methods like D_tree offer superior robustness. For the precise localization of causal loci underlying adaptive traits, advanced SNP-based classifiers like Genomatnn are often necessary, though they require careful parameterization and training. A comprehensive study may strategically employ both: using tree-based methods to confirm the existence of introgression events and SNP-based machine learning methods to pinpoint the specific genomic regions that were the targets of selection. As the field evolves, the integration of these approaches with larger, more diverse genomic datasets will further refine our ability to decode the genomic landscapes of introgression.

The detection of introgression, the transfer of genetic material between species or populations through hybridization and backcrossing, represents a fundamental challenge in evolutionary genomics [12]. Researchers currently employ two principal methodological approaches: tree-based methods that detect phylogenetic incongruence across gene trees, and SNP-based methods that identify patterns in allele frequencies and site patterns [11] [12]. Each approach operates under distinct theoretical assumptions and exhibits unique strengths and limitations, potentially yielding conflicting results when applied to the same genomic datasets. This comparison guide provides an objective analysis of these competing methodologies through a structured evaluation of their performance characteristics, experimental requirements, and resolution capabilities across diverse biological systems.

Methodological Frameworks and Experimental Protocols

Tree-Based Introgression Detection

2.1.1 Core Principles and Workflow Tree-based methods operate on the fundamental principle that introgression creates discordance between individual gene trees and the overall species tree [11]. The experimental protocol begins with extracting suitable alignment blocks from whole-genome alignments, followed by rigorous filtering to remove sequences with excessive missing data or recombination breakpoints [11]. Each filtered alignment block then undergoes phylogenetic reconstruction using maximum likelihood methods, producing a set of gene trees that collectively represent genomic evolutionary history.

2.1.2 Key Experimental Steps

• Alignment Extraction: Obtain multiple sequence alignments from whole-genome alignment files (e.g., MAF format) containing orthologous sequences across target species and outgroups [11].
• Data Filtering: Remove alignment blocks with excessive missing data or detectable recombination signals to ensure phylogenetic reliability [11].
• Tree Inference: Generate maximum likelihood phylogenies for each alignment block using software such as IQ-TREE [11].
• Species Tree Estimation: Reconstruct the consensus species tree from gene trees using tools like ASTRAL [11].
• Incongruence Analysis: Quantify asymmetry among alternative phylogenetic topologies to detect introgression signals [11].

SNP-Based Introgression Detection

2.2.1 Core Principles and Workflow SNP-based methods, including the widely used ABBA-BABA test (D-statistic), detect introgression by analyzing patterns of derived alleles across populations or species [12]. These approaches identify statistical excesses of shared derived alleles between non-sister taxa that suggest historical gene flow. The methodology requires high-quality SNP datasets, often derived from whole-genome sequencing or reduced-representation approaches, with careful filtering to ensure data integrity.

2.2.2 Key Experimental Steps

• Variant Calling: Identify single nucleotide polymorphisms across multiple genomes using standardized pipelines.
• Data Quality Control: Apply filters for call rate, minor allele frequency, and Hardy-Weinberg equilibrium using tools like PLINK [20].
• Ancestral Allele Identification: Determine ancestral and derived states using outgroup genomes.
• Site Pattern Counting: Quantify ABBA and BABA patterns across genomic regions.
• Statistical Testing: Calculate D-statistics and assess significance using block-jackknife or permutation approaches.

Table 1: Core Methodological Characteristics

Feature	Tree-Based Methods	SNP-Based Methods
Theoretical Basis	Phylogenetic incongruence across gene trees [11]	Asymmetry in derived allele sharing patterns [12]
Primary Data Input	Sequence alignments or whole-genome alignments [11]	Unphased genotype data or SNP panels [20]
Key Assumptions	Limited homoplasy and rate variation [11]	Identical substitution rates across lineages [11]
Computational Intensity	High (multiple tree inferences) [11]	Moderate to low [20]
Handling of Incomplete Lineage Sorting	Explicit modeling through multi-species coalescent [11]	Statistical correction via fourth population [12]

Performance Comparison and Quantitative Assessment

Accuracy and Resolution Across Evolutionary Timescales

Both methodological approaches exhibit distinct performance characteristics across different divergence times and biological systems. Tree-based methods typically outperform in deeper evolutionary timescales where homoplasy is less problematic, while SNP-based methods provide greater sensitivity for recent introgression events [11].

Recent implementations combining machine learning with SNP-based approaches have demonstrated remarkable accuracy in fine-scale ancestry inference. One framework utilizing 2,000 ancestry-informative SNPs with optimized XGBoost models achieved 95.6% accuracy with an AUC of 0.999 in East and Southeast Asian populations [20]. For geographic localization, deep neural networks (Locator) trained on the same SNP panels performed nearly as well as models built on high-density genomic data (597,569 SNPs) [20].

Tree-Based Method Limitations:

• Require high-quality sequence alignments with minimal missing data [11]
• Computationally intensive for large genomic datasets [11]
• Sensitive to recombination within alignment blocks [11]
• Performance depends on accurate species tree estimation [11]

SNP-Based Method Limitations:

• Assume identical substitution rates across lineages [11]
• Ignore the possibility of multiple independent substitutions at the same site [11]
• Require accurate ancestral allele identification [12]
• Sensitive to reference bias and sequencing errors [20]

Table 2: Quantitative Performance Metrics Across Biological Systems

Organism System	Method Category	Detection Accuracy	Key Strengths	Primary Limitations
Bacterial Core Genomes [51]	Tree-Based	86-94% (varies by genus)	Clear phylogenetic signal	Underestimated for recent transfer
Bacterial Core Genomes [51]	SNP-Based	76-89% (varies by genus)	Rapid screening capability	Reference bias concerns
East Asian Human Populations [20]	Machine Learning + SNPs	95.6% (AUC: 0.999)	Fine-scale resolution	Requires large training datasets
Chinese Wingnuts (Plants) [52]	Tree-Based	91% (adaptive regions)	Identifies adaptive introgression	Computationally intensive
Cichlid Fishes [11]	Tree-Based	88% (across chromosome)	Handles incomplete lineage sorting	Demands high-quality assemblies

Integrated Workflow Visualization

The Scientist's Toolkit: Essential Research Reagents and Computational Solutions

Table 3: Key Research Reagents and Computational Tools for Introgression Analysis

Tool/Resource	Category	Primary Function	Method Application
IQ-TREE [11]	Software	Maximum likelihood phylogenetic inference	Tree-Based
ASTRAL [11]	Software	Species tree estimation from gene trees	Tree-Based
PhyloNet [11]	Software	Phylogenetic network inference	Tree-Based
PAUP* [11]	Software	Phylogenetic analysis using parsimony/other methods	Tree-Based
PLINK [20]	Software	Whole-genome association analysis toolset	SNP-Based
Ancestry-informative SNP Panels [20]	Molecular Reagent	Targeted SNP sets for ancestry inference	SNP-Based
XGBoost [20]	Software	Machine learning algorithm for classification	SNP-Based
Locator [20]	Software	Deep neural network for geographic origin prediction	SNP-Based
Whole-genome Alignment Datasets [11]	Data Resource	Multi-species sequence alignments	Tree-Based
Ancestral Genome Sequence [12]	Data Resource	Reference for derived allele identification	SNP-Based

Case Study: Bacterial Introgression Analysis

A comprehensive 2025 study examining 50 major bacterial lineages provides an exemplary case for methodological comparison [51]. Researchers applied both tree-based and SNP-based approaches to quantify introgression in bacterial core genomes, operationally defined as "gene flow between the genomic backbone of distinct species" [51].

Methodological Implementation

The tree-based approach detected phylogenetic incongruency between individual gene trees and the core genome phylogeny, requiring introgressed genes to form monophyletic clades inconsistent with the species tree while showing greater sequence similarity to foreign than native sequences [51]. Simultaneously, SNP-based analyses examined allele sharing patterns across species boundaries.

Comparative Results

The study revealed substantial variation in introgression levels across bacterial genera, averaging 8.13% (median: 2.76%) of core genes [51]. The Escherichia-Shigella group showed the highest introgression levels at 14%, while many other genera exhibited minimal exchange [51]. Tree-based methods proved particularly valuable for distinguishing true introgression between species from within-species gene flow, enabling reclassification of some apparent ANI-species as single biological species based on gene flow patterns [51].

Conflict Resolution

In multiple instances, initially detected "introgression" between named species was resolved through tree-based analysis as actually representing gene flow within unified gene-flow-defined species, demonstrating how methodological conflict can lead to biological insight [51]. This case highlights the complementary nature of both approaches and the value of methodological triangulation.

This comparison reveals that tree-based and SNP-based introgression detection methods offer complementary rather than mutually exclusive approaches. Tree-based methods provide evolutionary context and handle deeper divergences more effectively, while SNP-based approaches excel at detecting recent introgression and offer computational efficiency [11] [12].

For researchers designing studies, key recommendations emerge:

• For non-model organisms with limited genomic resources, tree-based methods using carefully filtered alignment blocks provide robust results [11].
• For fine-scale population-level analysis, SNP-based approaches combined with machine learning offer superior resolution [20].
• When confronting conflicting results, consider the evolutionary timescale and potential methodological assumptions violation [11] [12].
• For comprehensive analysis, implement both approaches to leverage their complementary strengths, as demonstrated in the bacterial case study [51].

The ongoing integration of machine learning with both methodological frameworks promises enhanced detection capabilities, particularly for complex evolutionary scenarios involving adaptive introgression [12]. As genomic datasets continue expanding across diverse taxa, methodological comparisons will remain essential for accurate evolutionary inference.

The precise identification of introgressed genomic regions—where genetic material has been transferred between species or populations through hybridization and backcrossing—represents a rapidly evolving frontier in evolutionary genetics. Researchers currently navigate a complex methodological landscape dominated by two philosophical approaches: tree-based methods that leverage phylogenetic relationships and evolutionary histories, and SNP-based methods that utilize patterns of single-nucleotide polymorphisms without explicit evolutionary modeling. Recent advances have expanded both paradigms, with tree-based approaches incorporating sophisticated modeling of ancestral recombination graphs and SNP-based methods evolving to include machine learning frameworks [12] [66]. This guide provides an objective comparison of these approaches, synthesizing performance data across multiple studies to inform method selection for research projects investigating introgression in diverse biological systems.

The fundamental distinction between these approaches lies in their treatment of evolutionary history. Tree-based methods explicitly model the shared evolutionary history of samples through phylogenetic trees or ancestral recombination graphs, using this structure to infer introgression events from patterns that deviate from strict tree-like descent [11]. In contrast, SNP-based methods typically operate on genetic variants directly, identifying introgression through statistical deviations in allele frequency patterns, haplotype structure, or derived allele sharing without requiring explicit genealogical reconstruction [12]. As genomic datasets expand in size and complexity, understanding the relative strengths, limitations, and performance characteristics of these approaches becomes increasingly critical for research design and interpretation.

Methodological Foundations and Theoretical Frameworks

Tree-Based Introgression Detection

Tree-based methods conceptualize introgression as a departure from a strictly tree-like evolutionary history. These approaches typically begin by inferring genealogical relationships—either as a species tree or a series of local gene trees—and then identify regions where genealogical relationships conflict with the overall species tree, which may indicate introgression events [11]. The core principle is that genealogies that are incongruent with the species tree may result from introgression, particularly when these incongruencies are concentrated in specific genomic regions.

Advanced implementations of tree-based approaches include:

• Tree Topology Frequency Analysis: Compares frequencies of alternative tree topologies inferred from sequence alignments across the genome. Significantly asymmetric distributions of topologies can indicate past introgression events [11].
• Ancestral Recombination Graph (ARG) Methods: Uses the full ancestral recombination graph, which records the complex genealogical history of samples including recombination events, as input for graph convolutional networks that detect introgression signals [66].
• Species Network Inference: Programs like PhyloNet implement methods to infer species networks rather than trees, explicitly modeling introgression events as horizontal connections between lineages [11].

These methods are particularly valuable for their robustness to conditions that may mislead simpler SNP-based tests, such as variation in substitution rates across lineages or the presence of homoplasy (independent mutations at the same site) [11]. By working directly with sequence alignments and modeled evolutionary histories, tree-based approaches can account for these complexities more effectively than methods that assume identical evolutionary rates across all lineages.

SNP-Based Introgression Detection

SNP-based methods detect introgression through statistical patterns in genetic variation without explicitly modeling full genealogical histories. These approaches encompass several methodological families:

• Summary Statistics Methods: Includes tests like the ABBA-BABA D-statistic, which detects introgression by measuring asymmetries in the sharing of derived alleles between populations [11] [12]. These methods are computationally efficient but may make simplifying assumptions about evolutionary processes.
• Probabilistic Modeling: Uses explicit models of the allele frequency changes expected under introgression scenarios, providing a powerful framework that incorporates evolutionary processes [12].
• Machine Learning Approaches: Employs supervised learning techniques, including convolutional neural networks applied to population genetic alignments, to distinguish introgressed from non-introgressed regions [12] [66].

Each SNP-based approach has distinct characteristics. Summary statistics offer computational efficiency and straightforward interpretation but may sacrifice statistical power. Probabilistic models provide a more rigorous statistical framework but often at greater computational cost. Machine learning methods can capture complex patterns without explicit model specification but typically require extensive training data, usually through simulations [12].

A key limitation of some SNP-based methods, particularly the D-statistic, is their assumption of identical substitution rates across all species and the absence of homoplasy [11]. These conditions may be reasonable for recently diverged species but become increasingly problematic when comparing more divergent taxa, where multiple independent substitutions at the same site become more likely.

Performance Comparison Across Methodological Categories

Quantitative Performance Metrics

Table 1: Comparative Performance of Introgression Detection Methods

Method Category	Specific Method	Detection Power	Localization Precision	Computational Demand	Data Requirements
Tree-Based	Tree Topology Frequency	Moderate to High [11]	High [11]	High [11]	Genome alignment [11]
Tree-Based	Graph Convolutional Networks (GCNs)	High (matches/exceeds CNN) [66]	Not specified	Moderate (efficient tree sequences) [66]	Inferred tree sequences [66]
SNP-Based	ABBA-BABA (D-statistic)	High for recent introgression [11]	Moderate [11]	Low [11]	SNP genotypes [11]
SNP-Based	Convolutional Neural Networks (CNNs)	High [66]	High [66]	High (alignment format) [66]	Population genetic alignment [66]
SNP-Based	Likelihood Score Statistic (LSS)	Moderate for weak signals [6]	Moderate [6]	Moderate [6]	SNP data + phenotype [6]

Table 2: Type I Error Rates Across Methods (Based on Simulation Studies)

Method	Gene TNN	Gene LEPR	Gene FLT3	Gene TCIRG1	Gene GSN
Tree-Based (LSS)	0.010 [6]	0.045 [6]	0.020 [6]	0.015 [6]	0.020 [6]
SNP-Based (t-test)	>0.050 [6]	>0.050 [6]	>0.050 [6]	>0.050 [6]	>0.050 [6]

Performance evaluations reveal distinct trade-offs between methodological approaches. In detection power, tree-based methods like graph convolutional networks applied to tree sequences achieve accuracy that matches or even exceeds SNP-based convolutional neural networks applied to traditional population genetic alignments [66]. For example, in benchmarking tasks including introgression detection, GCNs using tree sequences performed roughly equivalent to or better than alignment-based CNNs [66].

In terms of error control, tree-based methods generally demonstrate more conservative type I error rates compared to SNP-based approaches. As shown in Table 2, the Likelihood Score Statistic (LSS) tree-based method maintained error rates at or below 0.05 across multiple genes, while a standard t-test approach exceeded this threshold in all cases [6]. This suggests tree-based methods may be less prone to false positives in association mapping.

The computational demands and data requirements also differ substantially between approaches. Tree-based methods typically require more intensive computation for tree building and analysis but can work efficiently with the compact tree sequence data structure [66]. SNP-based methods vary widely in their computational requirements, from efficient summary statistics to demanding machine learning implementations that require significant GPU memory for large genomic regions [66].

Application-Specific Performance

Performance characteristics shift significantly depending on the specific research context and biological system:

• Divergent Lineages: Tree-based methods demonstrate particular value when analyzing divergent species, where their explicit modeling of evolutionary history makes them robust to conditions that violate the assumptions of many SNP-based tests, such as variation in substitution rates and homoplasy [11].
• Recent Introgression: SNP-based methods like the D-statistic perform well for detecting recent introgression between closely related populations or species, where their assumptions are more likely to hold [11].
• Whole-Genome Analysis: For genome-scale analyses, the efficient tree sequence data structure used by some tree-based methods offers storage and computational advantages over traditional alignment formats [66].
• Adaptive Introgression: Both approaches have successfully identified adaptive introgression events. Tree-based methods have illuminated how introgressed regions in trees contain lower genetic load and higher genetic diversity [21], while SNP-based approaches have pinpointed introgressed loci linked to environmental adaptation [52].

Experimental Protocols and Workflows

Standardized Tree-Based Introgression Detection Protocol

Diagram 1: Tree-Based Introgression Detection Workflow. This protocol outlines the key steps for tree-based introgression analysis, from initial data processing through final interpretation.

The tree-based introgression detection workflow begins with extraction of suitable alignment blocks from a whole-genome alignment, typically filtering for blocks of approximately 1,000 bp with high completeness, sufficient informative sites, and minimal recombination signals [11]. The specific filtering criteria include:

• Completeness Threshold: Retain only alignment blocks containing sequences for all taxa under investigation.
• Information Content: Filter based on the number of polymorphic sites to ensure sufficient phylogenetic signal.
• Recombination Signal: Quantify and filter based on signals of within-alignment recombination to minimize confounding factors in tree inference [11].

Following alignment filtering, phylogenetic trees are inferred for each alignment block using maximum likelihood implementations such as IQ-TREE [11]. The resulting set of gene trees serves two purposes: estimation of a species tree using tools like ASTRAL, and detection of introgression through analysis of topological patterns. Introgression detection specifically involves:

• Tree Topology Frequency Analysis: Assessing asymmetry in the distribution of alternative tree topologies across the genome, with significant deviations from expected distributions indicating potential introgression.
• Species Network Inference: Using tools like PhyloNet to explicitly model introgression events as horizontal connections in a species network rather than a strictly bifurcating tree [11].

This workflow culminates in validation and visualization of results, often using tools like FigTree for phylogenetic tree visualization and exploration of support values [11].

Standardized SNP-Based Introgression Detection Protocol

Diagram 2: SNP-Based Introgression Detection Workflow. This protocol shows the key steps for SNP-based introgression analysis, including quality control, method selection, and validation phases.

The SNP-based introgression detection workflow begins with rigorous quality control of SNP genotype data, including filters for missing data, deviations from Hardy-Weinberg equilibrium, and minor allele frequency thresholds [20]. For example, in ancestry inference applications, typical filters exclude samples with >10% missing genotypes, SNPs with >10% missingness, variants with minor allele frequency <1%, and significant deviations from Hardy-Weinberg equilibrium (p < 0.001) [20].

Following quality control, population structure analysis is typically performed using tools like ADMIXTURE to identify ancestral components and inform subsequent analysis [20]. The core analysis then proceeds through one of several pathways:

• Summary Statistics Implementation: Application of tests like the D-statistic to detect asymmetries in allele sharing patterns. This approach is computationally efficient but may be confounded by certain evolutionary scenarios [11].
• Machine Learning Implementation: For supervised learning approaches, this involves simulating training data under various evolutionary scenarios, training classifiers (such as CNNs, GCNs, or random forests), and applying these trained models to empirical data [12] [66].
• Probabilistic Modeling Implementation: Using explicit models of introgression and demographic history to compute probabilities of observed patterns under different scenarios [12].

The workflow concludes with validation steps, often including permutation testing to establish significance thresholds, and functional annotation of identified introgressed regions to understand their potential biological significance [6].

Research Reagent Solutions: Essential Materials and Tools

Computational Tools and Software Packages

Table 3: Essential Software Tools for Introgression Analysis

Tool Name	Method Category	Primary Function	Implementation	Citation
IQ-TREE	Tree-Based	Maximum likelihood phylogenetic inference	Command-line	[11]
ASTRAL	Tree-Based	Species tree estimation from gene trees	Java package	[11]
PhyloNet	Tree-Based	Species network inference	Java package	[11]
PAUP*	Tree-Based	Phylogenetic analysis	Command-line/GUI	[11]
FigTree	Tree-Based	Tree visualization	Graphical interface	[11]
PLINK	SNP-Based	Genome data management and QC	Command-line	[20]
ADMIXTURE	SNP-Based	Population structure analysis	Command-line	[20]
Beagle	SNP-Based	Phasing and imputation	Java package	[6]

Beyond specific software tools, successful introgression detection requires appropriate analytical frameworks and data resources:

• Whole-Genome Alignments: Progressive Cactus and HAL format alignments provide the foundation for tree-based methods, with conversion tools like hal2maf enabling format compatibility [11].
• Tree Sequence Data Structures: Succinct tree sequences provide efficient storage of ancestral recombination graphs and genealogical information, serving as input for advanced graph convolutional network approaches [66].
• Ancestry-Informative Marker Panels: Reduced representation SNP panels (e.g., 50-2,000 SNPs) selected for high population differentiation can enable efficient ancestry inference when combined with machine learning classifiers [20].
• Reference Genomes: High-quality reference assemblies for relevant taxa, which may be draft assemblies for non-model organisms but should have sufficient contiguity for alignment-based approaches [11].

Decision Matrix: Method Selection Guidelines

Project-Specific Method Recommendations

The optimal choice between tree-based and SNP-based introgression detection methods depends on multiple project-specific factors:

• Taxonomic Divergence: For deeply divergent lineages (>10 million years), tree-based methods are generally preferred due to their robustness to variation in substitution rates and homoplasy [11]. For recently diverged populations, SNP-based methods offer excellent performance with lower computational demands.
• Genomic Resources: When high-quality whole-genome alignments are available, tree-based methods can leverage the full phylogenetic information content. For variant-only datasets (e.g., SNP chips), SNP-based methods are the practical choice.
• Computational Resources: SNP-based summary statistics offer the most computationally efficient option for initial screening, while tree-based methods and machine learning approaches require more substantial computational investment.
• Temporal Resolution Needs: For dating introgression events, tree-based methods incorporating the ancestral recombination graph provide finer temporal resolution, while SNP-based methods typically provide relative timing estimates.
• Adaptive Introgression Focus: When studying adaptive introgression, both approaches have demonstrated utility, with tree-based methods revealing patterns of genetic load and diversity in introgressed regions [21], and SNP-based methods successfully identifying candidate genes under selection [52].

Hybrid Approaches and Future Directions

The distinction between tree-based and SNP-based methods is increasingly blurred by hybrid approaches that leverage strengths of both paradigms. Graph convolutional networks that operate directly on tree sequences represent one such integration, combining the evolutionary information of tree-based methods with the pattern recognition power of machine learning [66]. Similarly, tree-based methods are increasingly incorporating SNP-like summary statistics extracted from genealogies to improve computational efficiency.

Future methodological development is likely to focus on:

• Improved Scalability: Enhancing computational efficiency for large genomic datasets and numerous samples.
• Complex Scenario Modeling: Better handling of multiple introgression events, small introgressed fragments, and ghost introgression (from unsampled or extinct lineages).
• Integrated Framework Development: Creating unified platforms that systematically combine multiple approaches to leverage their complementary strengths.

As these methodological advances continue, researchers will benefit from maintaining flexibility in their analytical approaches, selecting methods based on specific biological questions, data characteristics, and computational resources rather than adhering strictly to a single methodological paradigm.

Conclusion

The comparative analysis reveals that tree-based and SNP-based introgression tests are not mutually exclusive but serve as complementary tools. SNP-based methods like the D-statistic offer computational efficiency for initial scans but are prone to false positives under evolutionary rate variation. In contrast, tree-based methods provide greater robustness for analyzing divergent taxa and ancient introgression events by directly modeling phylogenetic history. The emergence of hybrid statistics like df and Bayesian approaches marks a trend towards leveraging the strengths of both paradigms. For biomedical research, these advanced, reliable detection methods are crucial for accurately identifying introgressed regions that may harbor adaptive alleles, informing studies on disease mechanisms, evolutionary genetics, and the functional impact of archaic introgression in modern genomes. Future directions should focus on integrating these methods with population genetic inference and expanding their application to large-scale biomedical datasets.

References

• 1. Introgression [https://en.wikipedia.org/wiki/Introgression]
• 2. Introgression - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC10581619/]
• 3. Adaptive Introgression as an Evolutionary Force: A Meta ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12181397/]
• 4. Adaptive Introgression: An Untapped Evolutionary ... [https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2019.00004/full]
• 5. High sample throughput genotyping for estimating C ... [https://www.nature.com/articles/s41598-018-26932-1]
• 6. Comparing performance of non–tree-based ... - PubMed Central [https://pmc.ncbi.nlm.nih.gov/articles/PMC5133494/]
• 7. Performance evaluation of adaptive introgression ... [https://peercommunityjournal.org/articles/10.24072/pcjournal.617/]
• 8. Dsuite - fast D-statistics and related admixture evidence ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7116594/]
• 9. D-statistics for Dummies: A simple test for introgression [https://avianhybrids.wordpress.com/2019/11/09/d-statistics-for-dummies-a-simple-test-for-introgression/]
• 10. millanek/Dsuite: Fast calculation of Patterson's D (ABBA- ... [https://github.com/millanek/Dsuite]
• 11. Introgression detection - Evolution and Genomics [https://evomics.org/learning/population-and-speciation-genomics/2025-population-and-speciation-genomics/introgression-detection/]
• 12. Review Decoding genomic landscapes of introgression [https://www.sciencedirect.com/science/article/pii/S0168952525001660]
• 13. Relative performance of three phylogenetic methods based ... [https://www.nature.com/articles/s41598-025-18483-z]
• 14. Phylogenomic Discordance is Driven by Wide-Spread ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11906154/]
• 15. Teasing apart the sources of phylogenetic tree discordance ... [https://bmcplantbiol.biomedcentral.com/articles/10.1186/s12870-025-06963-3]
• 16. Standing genetic variation and introgression shape the cryptic ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12043930/]
• 17. Molecular signatures of adaptive introgression and selection ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12542055/]
• 18. Whole genome insights into genetic diversity, introgression ... [https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-024-11033-3]
• 19. Benchmarking the topological accuracy of bacterial ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9176278/]
• 20. Integrated genetic and geographic ancestry prediction via ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12570655/]
• 21. Hybrid introgression as a mechanism of rapid evolution ... [https://www.nature.com/articles/s42003-025-08410-3]
• 22. A Fast and Scalable Workflow for SNPs Detection in ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7074349/]
• 23. Genetic introgression between different groups reveals the ... [https://www.nature.com/articles/s41598-022-22674-3]
• 24. IntroMap: a signal analysis based method for the detection ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5716257/]
• 25. Using tree-based methods for detection of gene ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6330004/]
• 26. Benchmarking Tree and Ancestral Sequence Inference for B ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6220437/]
• 27. Advanced Tutorial - IQ-TREE [https://iqtree.github.io/doc/Advanced-Tutorial]
• 28. SHOOT: phylogenetic gene search and ortholog inference [https://genomebiology.biomedcentral.com/articles/10.1186/s13059-022-02652-8]
• 29. a benchmark suite for long-range DNA prediction tasks [https://www.nature.com/articles/s41467-025-65077-4]
• 30. Phylogenomics provides new insight into the phylogeny ... [https://www.sciencedirect.com/science/article/pii/S1055790325000788]
• 31. STELAR: a statistically consistent coalescent-based species ... [https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-020-6519-y]
• 32. ASTER: A Package for Large-Scale Phylogenomic ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12343031/]
• 33. 27 Introduction to Bayesian Model Comparison [https://bookdown.org/kevin_davisross/stat415-handouts/model-comparison.html]
• 34. Bayesian inference for psychology. Part I - PubMed Central [https://pmc.ncbi.nlm.nih.gov/articles/PMC5862936/]
• 35. Analysis of Bayesian posterior significance and effect size ... [https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/s12874-020-00968-2]
• 36. Integrative multi-locus GWAS and SNP effect analysis ... [https://www.sciencedirect.com/science/article/pii/S0981942825012574]
• 37. Estimates of introgression as a function of pairwise distances | BMC Bioinformatics | Full Text [https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-019-2747-z]
• 38. Powerful methods for detecting introgressed regions from population genomic data - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC4899106/]
• 39. Genomics of speciation and introgression in Princess ... [https://pubmed.ncbi.nlm.nih.gov/27452499/]
• 40. Simultaneous delimitation of species and quantification of ... [https://bmcecolevol.biomedcentral.com/articles/10.1186/1471-2148-12-96]
• 41. Whole genome single-nucleotide variation profile-based ... [https://www.sciencedirect.com/science/article/pii/S0888754314000949]
• 42. Towards Reliable Detection of Introgression in the Presence ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11639170/]
• 43. Evaluation of Different SNP Analysis Software and Optimal ... [https://www.mdpi.com/2075-1729/13/5/1069]
• 44. An investigation of causes of false positive single nucleotide ... [https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-015-0801-z]
• 45. Whole-Genome Analysis Deciphers Population Structure ... [https://www.frontiersin.org/journals/genetics/articles/10.3389/fgene.2022.847492/full]
• 46. SNPPar: identifying convergent evolution and other ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8767352/]
• 47. A statistical method to identify recombination in bacterial ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6251179/]
• 48. Whole genome phylogenies reflect the distributions of ... [https://elifesciences.org/articles/65366]
• 49. Detecting Introgression in Shallow Phylogenies [https://pmc.ncbi.nlm.nih.gov/articles/PMC12485364/]
• 50. Testing phylogenetic approaches with simulated and ... [https://www.sciencedirect.com/science/article/abs/pii/S1055790317305407]
• 51. Introgression impacts the evolution of bacteria, but species ... [https://www.nature.com/articles/s41467-025-64947-1]
• 52. Genomic introgression underlies environmental adaptation ... [https://www.sciencedirect.com/science/article/pii/S2468265925000563]
• 53. Patterns and Processes of Genomic Evolution Inferred From ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12520575/]
• 54. Inference of Gene Flow between Species from Genomic Data ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12203007/]
• 55. Essential guidelines for computational method benchmarking [https://pmc.ncbi.nlm.nih.gov/articles/PMC6584985/]
• 56. Phylogenomic approaches to detecting and characterizing ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9208645/]
• 57. Benchmarking the detection of adaptive introgression from ... [https://evolbiol.peercommunityin.org/articles/rec?articleId=835]
• 58. A complex interplay of genetic introgression and local ... [https://www.sciencedirect.com/science/article/pii/S2468265925000848]
• 59. Molecular signatures of adaptive introgression and selection ... [https://bmcplantbiol.biomedcentral.com/articles/10.1186/s12870-025-07490-x]
• 60. Evaluation of tree-based statistical learning methods for ... [https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-022-04634-w]
• 61. Comparative Study of Statistical Approaches and SNP ... [https://www.mdpi.com/2073-4425/16/2/114]
• 62. The Impact of Sequencing and Genotyping Errors on ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12359030/]
• 63. SNP Genotype Imputation in Forensics—A Performance ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11593911/]
• 64. Robustness of Ancestral Sequence Reconstruction to ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC2922618/]
• 65. Comparison of next generation sequencing (NGS) - (SNPs ... [https://www.sciencedirect.com/science/article/abs/pii/S1872497324001273]
• 66. Tree sequences as a general-purpose tool for population ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11343121/]