Statistical properties of the methods for detecting positively selected amino acid sites

Parsimony and Bayesian methods have been developed for detecting positively selected amino acid sites. It has been reported that the parsimony method is generally conservative. In contrast, the Bayesian method is known to identify more positively selected sites than the parsimony method, especially when the number of sequences analyzed is small, although the interpretation of results obtained from the former method is controversial. Here I show that the likelihood-ratio test (LRT) of the Bayesian method corresponds to the parsimony method with window analysis, by analyzing the nucleotide sequences encoding the trans-activator (tax) gene of human T-cell lymphotropic virus type I (HTLV-I). It is also indicated that in the parsimony method, the test of selective neutrality using the binomial probability tends to be conservative, but the Monte Carlo simulation is useful for solving this problem. In addition, in the Bayesian method, the bootstrap method appears to produce similar results to the LRT. This information may be useful for improving the methods for detecting positively selected amino acid sites.