Modeling evolution of protein coding DNA sequences

Abstract

We develop a new class of computationally feasible stochastic models for statistical analysis of genetic sequence evolution and inference of properties of the underlying substitution processes in the context of maximum likelihood framework. Existing models for evolution of protein coding sequences allow site to site variation in non-synonymous substitution rates, but assume that the rate of synonymous substitutions is constant for all sites. New models provide a rigorous statistical framework for testing the hypothesis of synonymous rate constancy, and enable a host of data exploration and analysis tools. For several indicative data sets, the constancy assumption is shown to be violated, and some possible explanations are given. We also present an algorithm for improving efficiency of maximum likelihood evaluations, and discuss HyPhy--a user friendly and publicly distributed software implementation of our methods.