Robust estimation of natural selection using parametric codon models

This chapter focuses on the use of parametric codon models to test the mode of natural selection. It describes the nature of the different models, demonstrating the importance of model assumptions to reliably estimating selective neutrality. It examines the relationship between context-dependent processes on non-overlapping units and those relevant to codon evolutionary processes. To illustrate theoretical properties, the chapter presents both simple simulations and analyses of neutrally evolving DNA sequences from primates. It shows that estimates of natural selection using the commonly employed codon models of Muse and Gaut (1994), and of Goldman and Yang (1994), can both be strongly biased by sequence composition. It further illustrates that the Goldman and Yang model can have counterintuitive behaviour in hierarchical hypothesis-testing contexts.