Probabilistic Phylogenetic Inference with Insertions and Deletions
Figure 2
Comparison of dnaml versus dnamlε for ungapped alignments.
Tree reconstruction for ungapped alignments generated according to a F84 substitution model (2.0 transition to transversion ratio and equiprobable residues), for nine different time divergences, ranging from 0.005 to 2.0 substitutions per site and branch. For a given divergence value, 100 random trees with eight taxa were used. For each tree, single alignments were generated with lengths ranging from 50 to 1000 residues in 5 residue increments. For each alignment, a tree was inferred using the programs dnaml and dnamlε. Results are displayed as a function of the length of the alignments. (A) Fraction of trees which topology was correctly inferred as a function of the alignment length. The best performance occurs for alignments that contain about 18% pairwise substitutions on average (0.05 substitutions per site and branch). In this case, detectability seems to asymptote to approximately 78% for alignments of at least 800 residues. (B,C) Corresponding results when using the SDD and nBSD measures respectively. (D) Average mean branch length for each length bin. Overall, the two methods show similar performance for ungapped alignment. We mark with an arrow some extreme cases in which the two methods perform differently when inferring the tree branch lengths. (E) Comparison of computational time performance.