Assessment of substitution model adequacy using frequentist and Bayesian methods

Area of Science:

• Evolutionary Biology
• Computational Biology
• Phylogenetics

Background:

• Model-based phylogenetic methods (Maximum Likelihood, Bayesian) require appropriate molecular evolution models.
• Model selection methods (e.g., Likelihood Ratio Test, AIC) are common but may not reject inadequate models.
• Model adequacy tests (e.g., Goldman-Cox, Posterior Predictive Simulations) assess absolute model fit.

Purpose of the Study:

• Evaluate common substitution model adequacy using frequentist and Bayesian methods.
• Compare adequacy testing with standard model selection techniques.
• Investigate the impact of model adequacy on phylogenetic inference (topology, branch lengths, bipartition support).

Main Methods:

• Empirical and simulated data were used.
• Frequentist (Goldman-Cox) and Bayesian (Posterior Predictive Simulations) adequacy tests were applied.
• Model adequacy was compared against model selection criteria.
• Performance was assessed in Maximum Likelihood and Bayesian phylogenetic analyses.

Main Results:

• Tests of adequacy often failed to reject simple models, especially with among-site rate variation (ASRV).
• Adequacy tests tended to select simpler models than model selection methods.
• Using simplest adequate models yielded similar but divergent tree topologies and branch lengths.
• Bipartition support estimates could be affected, particularly for poorly supported nodes.
• ASRV assumptions influenced tree topology, length, and support.

Conclusions:

• The simplest adequate substitution models may serve as a viable alternative to complex models identified by selection methods.
• Model adequacy is important, but challenges in tree shape estimation can still impact phylogenetic inference.
• Careful consideration of model adequacy and assumptions (like ASRV) is vital for robust phylogenetic results.