Looking for Darwin in Genomic Sequences: Validity and Success Depends on the Relationship Between Model and Data

Area of Science:

• Evolutionary biology
• Computational biology
• Population genetics

Background:

• Codon substitution models (CSMs) are widely used with maximum likelihood (ML) to detect positive Darwinian selection.
• Concerns exist regarding the reliability of CSMs, with claims of frequent false conclusions.

Purpose of the Study:

• To identify and explain four statistical challenges in CSM-based evolutionary inference.
• To assess the limitations of CSMs in extracting historical information from sequence data.
• To propose methods for more robust inference of evolutionary mechanisms.

Main Methods:

• Case study approach to identify statistical difficulties.
• Overview of codon evolution from a population genetics perspective.
• Novel presentation of the ML statistical framework.
• In silico experiments using the mutation-selection modeling framework (MutSel).

Main Results:

• Identified four key statistical issues: model misspecification, low information content, confounding of processes, and phenomenological load (PL).
• Demonstrated that increased model complexity in later CSM development phases can hinder accurate inference.
• Highlighted the importance of the relationship between model and data for valid conclusions.

Conclusions:

• CSMs face legitimate statistical challenges that can lead to inaccurate evolutionary inference.
• Appropriate model selection, informed by the data, is essential to avoid these problems.
• In silico experiments and additional analyses like penalized LRTs and PL estimation are vital for reliable evolutionary studies.