Long-Branch Attraction Bias and Inconsistency in Bayesian Phylogenetics
Datasets of 5,000 nt were generated using heterogeneous evolutionary models on a four-taxon tree with non-sister long terminal (0.75 substitutions/site) and short terminal (0.05) branch lengths and an internal branch of 0.02, then analyzed using a simple homogeneous model. We plotted the proportion of replicates from which each topology was recovered, as well as the mean posterior probability of each tree, as evolutionary heterogeneity increased. a, Sequences were generated with convergent G+C content in non-sister lineages. GC heterogeneity indicates absolute increase of G+C content in the marked lineages above ancestral baseline of 30%. b, Two classes of heterotachous sites evolved on the same topology but with different branch lengths for each class. We varied the strength of heterogeneity by increasing from zero the proportion of sites in the first class.