Testing adequacy of codon substitution models

Abstract

In phylogenetic inference, codon substitution models are mainly used to detect positive selection, which is a sign of adaptive molecular evolution at the protein level. Positive selection is identified when non-synonymous substitutions are more frequent than synonymous ones. To model the evolution of amino acid and codon sequences, Markov chains can be used.

It's important to test these models for adequacy before drawing phylogenetic conclusions, as inadequate models can lead to unreliable results and incorrect biological interpretations. This thesis introduces several methods to evaluate the adequacy of codon substitution models, such as Pearson's Chisq test with two alternative strategies for binning site patterns; influence matrix based binning and random binning; and the Anderson-Darling test. These methods help determine whether the proposed model effectively fits the data, thereby assessing the reliability of conclusions derived from it.