Estimation of Cross-Species Introgression Rates Using Genomic Data Despite Model Unidentifiability

Area of Science:

• Evolutionary biology
• Genomics
• Computational phylogenetics

Background:

• Multispecies coalescent with introgression (MSci) models are crucial for inferring species divergence and gene flow from genomic data.
• These models are known to suffer from unidentifiability issues, where different parameters or models yield indistinguishable predictions.
• Previous unidentifiability studies relied on gene trees, inefficiently utilizing genomic data.

Purpose of the Study:

• To investigate and characterize unidentifiability in MSci models, specifically focusing on the bidirectional introgression (BDI) model under full-likelihood methods.
• To develop novel algorithms for processing Markov chain Monte Carlo (MCMC) samples to address label-switching problems inherent in MSci models.
• To provide guidelines for utilizing MSci models effectively for inferring gene flow from genomic data.

Main Methods:

• Characterization of unidentifiability in bidirectional introgression (BDI) models using full-likelihood approaches.
• Derivation of rules for label-switching unidentifiability in MSci models with multiple introgression events.
• Development and implementation of new algorithms for MCMC sample processing to resolve label-switching issues within the bpp software.

Main Results:

• Identified that MSci models with k bidirectional introgression events exhibit 2k unidentifiable posterior modes.
• Distinguished between within-model parameter unidentifiability (sister species introgression) and between-model unidentifiability (non-sister species introgression).
• Demonstrated the utility of the new algorithms in analyzing both synthetic and real genomic data.

Conclusions:

• The study provides a theoretical framework and practical tools to overcome unidentifiability challenges in MSci models.
• Novel algorithms effectively resolve label-switching problems, improving the reliability of gene flow inference.
• The findings offer crucial guidance for researchers using genomic data to reconstruct evolutionary histories involving introgression.