Scaling up Bayesian population phylogenomics through virtual dimension reduction.

Area of Science:

• Population phylogenomics
• Computational biology
• Evolutionary genetics

Background:

• Population phylogenomics jointly infers population genetic processes and phylogenetic trees.
• Bayesian inference under the multispecies coalescent (MSC) model using Markov chain Monte Carlo (MCMC) is the standard approach.
• MCMC methods face challenges with high-dimensional state spaces and lack of parallelization.

Purpose of the Study:

• To develop a novel computational method addressing MCMC limitations in population phylogenomics.
• To improve the efficiency and scalability of phylogenetic and population genetic inference.

Main Methods:

• Development of Virtual Dimension Reduction allowing Parallelisation (VDRoP).
• VDRoP achieves efficient MCMC mixing through analytical parameter integration without restricting priors.
• Implementation within the Bayesian Program for Phylogenetics (BPP) software.

Main Results:

• VDRoP significantly reduces MCMC run-times (3-8 fold) on empirical genomic datasets.
• The method demonstrates substantial improvements in mixing efficiency (up to 50 fold).
• Successful application to diverse datasets including Adansonia, Anopheles, and Heliconius.

Conclusions:

• VDRoP offers a computationally efficient and parallelizable alternative to standard MCMC for population phylogenomics.
• The method enhances the feasibility of analyzing large-scale genomic data for evolutionary inference.
• This advancement facilitates more robust reconstruction of species divergence and demographic histories.