FIG. 3.—

Mixed branch length model improves phylogenetic accuracy under simulated conditions derived from observations of heterotachy in empirical sequence data. Proportion of replicate data sets from which the correct tree was uniquely recovered by each method is plotted against increasing internal branch length; sequences of 5,000 nt were simulated using the trees at left, with long terminal branch lengths of 0.75 expected substitutions/site. (A) Sequences were simulated along a single set of branch lengths using a stationary covarion process in which sites switch between variable and invariable states as they evolve. Series are not shown for the true model (which is the same as the covarion model in this case) and the mixed branch length model (which is the same as the homotachous model). (B) Data were simulated under a model of covarion rate shifts in which some invariable sites become released from selection in nonsister lineages; other sites that were previously variable then become invariable to compensate for the relaxation of selective constraint at correlated sites. The proportion of correct inferences using each method is plotted against the length of the internal branch (for variable sites only) on the true tree. (C) Sequence data were simulated with convergent proportions of invariable sites in nonsister lineages.