Relative Model Fit Does Not Predict Topological Accuracy in Single-Gene Protein Phylogenetics

Area of Science:

• Phylogenetics
• Computational Biology
• Evolutionary Biology

Background:

• Phylogenetic reconstruction relies on selecting appropriate evolutionary models.
• Model selection typically uses information-theoretic criteria to rank model fit.
• It's assumed better model fit enhances phylogenetic accuracy.

Purpose of the Study:

• To investigate the relationship between model fit and topological accuracy in protein phylogenetics.
• To assess the impact of model misspecification on phylogenetic inference.
• To evaluate the utility of relative model selection for protein evolutionary analyses.

Main Methods:

• Utilized simulations with site-heterogeneous codon models, distinct from protein-level models.
• Analyzed real protein sequences to assess phylogenetic inference across various models.
• Compared topologies and false-positive rates inferred by models with different goodness-of-fit scores.

Main Results:

• Phylogenetic trees showed high topological consistency despite significant differences in model fit.
• All tested models inferred similar proportions of false-positive splits, indicating potential widespread model misspecification.
• The general time reversible (GTR) model performed comparably to models with fixed exchangeabilities.

Conclusions:

• Relative model selection may not hinder protein phylogenetic analysis but offers no predictable accuracy improvements.
• Model fit is not a reliable indicator of phylogenetic inference accuracy in protein evolution.
• Current protein evolution models might be fundamentally misspecified for real sequence data.