Figure 2

Comparison of the performance of bipartiton and quartet-based analyses. Increasing taxon sampling justifiably is expected to increase the reliability of phylogenetic reconstruction; however, the support values for bipartitions that include all taxa tends to drop as more taxa are added. Panel A depicts the phylogenies used for simulations. Starting with an unrooted tree of four leaves, ((A B),(C D)) and an internal branch of 0.01 average substitutions per site, we added 1, 4, 9, 19 and 49 additional leaves to the internal branch. Simulations for each topology were performed with Seq-Gen. [16] using the indicated trees, the WAG substitution matrix [17] and a Γ distribution with a shape parameter of 1 approximated by four discrete rate categories for the rate distribution. SEQBOOT from the PHYLIP package [18] was used to generate 100 bootstrap sequences and trees were reconstructed from each bootstrap sample using FastTree 2.1 [19] using the same model for sequence evolution and parameters “-spr 4”, “-mlacc 2”, and “-slownni” for increased reconstruction accuracy. For each topology the evolution of sequences of varying lengths (200, 500 and 1000 amino acids) was simulated. For each of the simulated data sets, we generated 100 bootstrap replicates and recorded the maximum support for a bipartition separating (A B) from (C D) (Panel B) and the bootstrap support for the embedded quartet ((A B),(C D)) for all simulations (Panel C). Error bars give the standard error of the mean from 100 replicates each