Improved inference of site-specific positive selection under a generalized parametric codon model when there are...

Area of Science:

• Evolutionary biology
• Molecular evolution
• Bioinformatics

Background:

• Detecting positive Darwinian selection in proteins typically involves estimating the nonsynonymous to synonymous substitution rate ratio (ω = dN/dS).
• Traditional codon models often simplify the substitution process by assuming uniform nonsynonymous rates and ignoring instantaneous double/triple nucleotide mutations.
• These simplifications can lead to inaccurate estimates of ω and impact the reliability of selection tests.

Purpose of the Study:

• To develop a flexible codon modeling framework that accommodates a more realistic substitution process.
• To implement likelihood ratio tests (LRTs) for positive selection (ω > 1) within this new framework.
• To evaluate the impact of modeling multiple nonsynonymous rates (MNRs) and instantaneous double/triple (DT) nucleotide mutations on selection inference.

Main Methods:

• Developed a general-purpose parametric (GPP) modeling framework for codons.
• The GPP model allows specification of all possible instantaneous codon substitutions, including MNRs and instantaneous DT nucleotide changes.
• Implemented LRTs for ω > 1 using GPP models and compared performance against traditional models (M2a, M8) via simulations and real data analysis.

Main Results:

• Failure to model MNRs and DT mutations can reduce statistical power and inflate false positive rates in selection detection.
• Traditional models showed high sensitivity to DT mutations, particularly under certain frequency parameterizations (e.g., MG).
• The GPP framework, by incorporating MNRs and DT mutations, significantly improved accuracy and power, though over-parameterization could degrade performance.

Conclusions:

• GPP models should be used in conjunction with traditional codon models for robust evolutionary inference.
• Experimental designs should include assessing model assumption robustness and investigating non-standard behavior of maximum likelihood estimates (MLEs).
• Further research is needed on model selection methods that balance model fit with the impact of un-modeled evolutionary processes.