Abstract
Recent advances in the estimation of amino acid exchangeability matrices have greatly improved the estimation of phylogenies. Here, we present an exchangeability matrix that specifically improves phylogenetic analysis of transmembrane proteins. We first use Bayesian inference to estimate replacement matrices for single families of transmembrane proteins. From these, we constructed a general replacement matrix, representative of all α-helical transmembrane proteins, in which the entries are the weighted harmonic means of the corresponding entries in the single-family replacement matrices. The best matrix improves the log-likelihood of transmembrane protein phylogenies by 0.002 points per site over a phylogeny made using the LG replacement matrix. Phylogenetic analysis of soluble proteins greatly benefits from the use of multiple exchangeability matrices that correspond to different structural features. The software that performs this analysis depends upon annotated sequence alignment for input. Accordingly, we also present a means of automatically an- notating individual residues of transmembrane protein structures. Using our structurally annotated alignments will allow construction of structure-specific exchangeability matrices for TM proteins, providing a means to incorporate structural information into phylogenetic inference of transmembrane proteins.