Infinite Mixture Models for Improved Modeling of Across-Site Evolutionary Variation

Area of Science:

• Evolutionary biology
• Computational biology
• Bioinformatics

Background:

• Phylogenetic trees are essential for biological studies using molecular sequence data.
• Accurate phylogenetic inference relies on accounting for evolutionary rate heterogeneity across sites.
• Current methods for partitioning sequence alignments into substitution models can be challenging.

Purpose of the Study:

• To develop and evaluate flexible Bayesian infinite mixture models for inferring across-site evolutionary variation.
• To simultaneously infer the number of partitions, site assignments, and evolutionary parameters.
• To explore novel approaches including hierarchical models and infinite hidden Markov models.

Main Methods:

• Utilized Bayesian infinite mixture models, including Dirichlet process mixtures.
• Developed and applied hierarchical models for grouped data structures.
• Implemented infinite hidden Markov models to capture spatial patterns in alignments.
• Adapted Markov chain Monte Carlo algorithms and parallel computing for scalability.

Main Results:

• Different mixture models show varying performance depending on the data scenario.
• Infinite hidden Markov models demonstrate particular promise for large datasets and complex evolutionary patterns.
• The framework was successfully implemented in the BEAST X software package.

Conclusions:

• Bayesian infinite mixture models offer a flexible and powerful approach to modeling evolutionary heterogeneity.
• Infinite hidden Markov models represent a significant advancement for phylogenetic inference, especially with complex genomic data.
• The developed framework enhances the accuracy and scalability of phylogenetic analyses in computational biology.