Long-Branch Attraction Bias and Inconsistency in Bayesian Phylogenetics
a, The correct eukaryote phylogeny places the microsporidian Encephalitozoon cuniculi with the fungi, as shown. The long branch attraction (LBA) tree pairs taxa in bold. b, We analyzed elongation factor-1 data using three evolutionary models: 1) JTT+G, Jones-Taylor-Thornton model of amino acid replacements with gamma-distributed among-site rate variation; 2) JTT+G:6, heterotachous mixture model with 6 branch-length classes, and 3) a 6-category partitioned model, with partitions inferred using JTT+G:6. For each model, we plot the log Bayes factor of the correct placement of microsporidia vs. the LBA tree, with positive values indicating support for the correct phylogeny (see Methods). Label ‘inf’ indicates maximal support for the LBA clade; the correct tree was not sampled during the MCMC run. c, We simulated 200 replicate sequence alignments of 500 residues along the tree in panel a, with branch lengths and model parameters estimated from elongation factor-1 data. Models used to simulate and analyze datasets are indicated in the figure. For each combination of models, we plotted the posterior probability of the incorrect LBA clade; bars indicate standard error.