OBAMA: OBAMA for Bayesian amino-acid model averaging

Area of Science:

• Phylogenetics
• Computational Biology
• Evolutionary Biology

Background:

• Bayesian phylogenetic analyses are widely used for amino acid alignments.
• Model selection for substitution and site models is crucial but often ad hoc.
• Existing methods for model selection do not account for overall model uncertainty.

Purpose of the Study:

• To introduce a novel method, OBAMA, for averaging over substitution and site models in phylogenetic analyses.
• To allow data to inform model choices and explicitly incorporate model uncertainty.
• To reduce bias in phylogenetic estimates by accounting for uncertainty in model selection.

Main Methods:

• The OBAMA method employs trans-dimensional Markov Chain Monte Carlo (MCMC) to switch between empirical amino acid substitution models (e.g., Dayhoff, WAG, JTT).
• It integrates switching between model-based or alignment-estimated base frequencies.
• The method also incorporates switching between gamma rate heterogeneity and proportion of invariable sites.

Main Results:

• A simulation study demonstrated the effectiveness of the OBAMA method.
• The method successfully estimates parameters like the proportion of invariable sites and the gamma shape parameter using appropriate priors.
• OBAMA reduces bias in phylogenetic estimates by accounting for model uncertainty.

Conclusions:

• The OBAMA method provides a robust framework for Bayesian phylogenetic inference by integrating model uncertainty.
• Implementation in the open-source OBAMA package for BEAST 2 facilitates joint tree inference across diverse models.
• This approach enhances the accuracy and reliability of phylogenetic reconstructions.