FIG. 2.—

Mixed branch length model improves phylogenetic accuracy under idealized heterotachous conditions. Sequences of 5,000 nt each were simulated using the tree at left in each panel, with long terminal branch lengths 0.75 substitutions/site and short terminals 0.05. Left graph in each panel plots the proportion of correct inferences made using homotachous, heterotachous, and the true partitioned model (see key at lower right) against the internal branch length of the true tree. Right graphs plot internal (left) and terminal (right) branch lengths estimated by maximum likelihood using true, mixed branch length, and homotachous models against the true internal branch length on which sequences were simulated; dotted lines indicate perfect correspondence between estimated and true lengths. (A) Felsenstein-zone heterotachy. (B) Inverse Felsenstein-zone heterotachy. Inset bar graph shows the proportion of replicates from which each method recovered each possible resolved topology when data were generated with an internal branch length of zero. (C) Signal-noise heterotachy. (D) Single long-branch heterotachy. Bottom left panel shows per-site likelihood calculated on an ideal infinite data set (see Materials and Methods) using homotachous ML plotted against increasing internal branch length; sequence was generated under SLBH conditions with a true internal branch length of 0.01. Bottom right panel shows accuracy when >95% support is required to resolve the phylogeny.