Methods for selecting fixed-effect models for heterogeneous codon evolution, with comments on their application to...

Area of Science:

• Evolutionary Biology
• Computational Biology
• Genomics

Background:

• Codon evolution models are crucial for understanding natural selection.
• Fixed-effect models incorporate biological knowledge to capture heterogeneous evolutionary dynamics across codon sites.
• These models partition sites based on biological factors like protein structure or gene function.

Purpose of the Study:

• To implement and evaluate fixed-effect codon models allowing for heterogeneity in substitution processes.
• To compare model selection strategies including backward elimination, AIC, and AICc.
• To provide recommendations for applying these models in real data analyses.

Main Methods:

• Implementation of fixed-effect codon models with partitioned substitution processes.
• Model selection strategies: backward elimination, Akaike Information Criterion (AIC), and corrected AIC (AICc).
• Performance evaluation through simulation studies and application to real datasets (abalone sperm lysin, Listeria flagellar proteins).

Main Results:

• Backward elimination demonstrated reliable performance for model selection in simulated data.
• Fixed-effect models were successfully applied to both single-gene (tertiary structure) and multi-gene (functional category) datasets.
• The study identified advantages and disadvantages of fixed-effect models, highlighting the need for a priori knowledge.

Conclusions:

• Backward elimination is recommended over AIC or AICc for model selection, with a stringent p-value cutoff (e.g., 0.0001).
• Sensitivity analysis is advised for robust results.
• Fixed-effect codon models are valuable tools for large-scale multi-gene analyses when applied thoughtfully.