Estimating Selection and Mutation Rates from a Random Field Model for Polymorphic Sites

Selection and mutation rates are estimated from aligned DNA sequences by using a Poisson random field model for base frequencies at polymorphic sites. This approach is applied at amino acid polymorphic codon positions to estimate selection rates against observed amino acid polymorphisms and also the proportion of codon positions that admit a weakly selected replacement. The technique can also be used at silent sites to obtain a numerical estimate of codon bias, which is shown in two examples to be of comparable strength to the selection against observed variant amino acids. A nonparametric test that can detect selection against amino acids using silent sites as a control is also discussed. Finally, a technique that estimates silent mutation rates assuming selective neutrality of silent polymorphisms, but which is not sensitive to saturation, is used as a consistency check.