Comparison of empirical model fits using BIC.

The best model (with smallest BIC) is shown in boldface. The BIC for , the model in which amino acid substitution pairs are randomly assigned to one of 5 rate classes, is estimated as the mean BIC over 100 permutations of rate class assignment. is the number of rate classes identified using a Genetic Algorithm model fitting procedure described in [20] and shown in parenthesis for each alignment. All models were fitted using maximum likelihood estimates of position-specific nucleotide frequencies.