Bayesian comparisons of codon substitution models.

Area of Science:

• Computational evolutionary biology
• Molecular evolution
• Bioinformatics

Background:

• Evolutionary models by Muse & Gaut (MG) and Goldman & Yang (GY) analyze nucleotide sequences using Markovian processes.
• Existing models, while foundational, lack a unified framework for comparing variations and extensions.
• Need for flexible approaches to account for complexities like heterogeneous rates and selection pressures.

Purpose of the Study:

• To develop and apply a general framework for contrasting MG and GY style codon substitution models.
• To evaluate recent extensions incorporating site-specific rates and selection effects.
• To assess the relative performance of different evolutionary models using Bayes factors.

Main Methods:

• Computation of Bayes factors to compare various codon substitution models.
• Inclusion of models with heterogeneous nonsynonymous rates and selection pressures.
• Application of posterior predictive checks to validate model parameterization.

Main Results:

• Results on three real datasets favor a logical model construction following the MG formulation.
• Demonstrated the importance of flexible parameterization for global amino acid or codon preferences.
• Highlighted the need for distinct parameters for nucleotide propensities alongside other evolutionary factors.

Conclusions:

• The proposed framework supports a broad modeling project in the MG style for evolutionary sequence analysis.
• Emphasizes combining and contrasting diverse model formulations for robust evolutionary inference.
• Stresses the importance of a sound probabilistic paradigm for developing and validating evolutionary models.