Stochastic Variational Inference for Bayesian Phylogenetics: A Case of CAT Model.

Area of Science:

• Genomics
• Molecular Evolution
• Computational Biology

Background:

• Molecular evolution patterns differ across genomic sites and genes.
• Bayesian infinite mixture models improve phylogenetic inference by accounting for evolutionary heterogeneity.
• Markov chain Monte Carlo (MCMC) methods are computationally intensive for large datasets.

Purpose of the Study:

• To develop a computationally efficient method for phylogenetic inference in genomics.
• To accelerate the PhyloBayes MPI program using a variational Bayesian approach.
• To address the computational limitations of MCMC sampling in analyzing complex evolutionary models.

Main Methods:

• Developed a variational Bayesian procedure to approximate posterior distributions.
• Implemented the method within the PhyloBayes MPI program to handle amino acid profile heterogeneity.
• Estimated variational distribution parameters by minimizing Kullback-Leibler divergence.

Main Results:

• The variational Bayesian method significantly reduced computational time (orders of magnitude) compared to MCMC.
• Accurately approximated phylogenetic trees, mixture proportions, and amino acid propensities.
• Demonstrated effectiveness on empirical datasets of mitochondrial, plastid-encoded, and nuclear proteins.

Conclusions:

• The variational Bayesian approach offers a computationally feasible alternative for robust phylogenetic inference.
• This method enables efficient analysis of complex genomic evolutionary processes with large datasets.
• Accelerated phylogenetic analysis can advance our understanding of molecular evolution across different genomic elements.