Relating physicochemical properties of amino acids to variable nucleotide substitution patterns among sites.

Markov-process models of codon substitution were implemented that account for features of DNA sequence evolution (such as transition/transversion bias and codon usage bias) as well as heterogeneity of amino acid substitution pattern over sites. The codon (amino acid) sites are assumed to come from several classes (such as secondary structure categories), among which the rate of amino acid substitution and the effect of amino acid chemical properties vary. Parameters are estimated by the maximum likelihood method, which accounts for the phylogenetic relationship among species and corrects for multiple hits at the same site. The likelihood ratio test is used to compare models. Mitochondrial cytochrome b genes of 28 primate species are analyzed. The site-heterogeneity models provide much better fit to previous homogeneous models.