Table 1.
Topological accuracy of trees inferred from simulated alignments.
Figure 1.
Local support values for splits found by PhyML with SPR moves and/or FastTree.
We examined local support values for the splits inferred by PhyML 3.0 with + SPRs on simulated alignments with 250 protein sequences. We classified PhyML's splits as correct and found by both PhyML and FastTree, correct but missed by FastTree, or incorrect. We show the distribution of support values for each class. The right-most bin includes the strongly supported splits (0.95 to 1.0), and the gray dashed line shows the uniform distribution. The support values are PhyML's minimum of the approximate likelihood ratio test [21] and SH-like [16], [17] local supports.
Table 2.
Average log-likelihood for genuine alignments with 500 sequences.
Table 3.
Comparison of RAxML and FastTree's log likelihoods, and the agreement of FastTree with RAxML's well-supported splits, for large genuine alignments.
Table 4.
Running time and memory usage on genuine alignments.
Figure 2.
Likelihoods over time for genuine alignments.
Each line shows the time it takes a different tool to reach a given likelihood. For the COG alignments, all times and likelihoods are averages over the seven alignments. For FastTree, we show the time and the improvement in likelihood for the minimum-evolution topology and the final (approximately-ML) topology. For RAxML, we show the maximum parsimony starting topology, the first two rounds of SPR moves, and the final topology (note the interrupted axis). For RAxML with FastTree's (minimum-evolution) starting tree, we show the starting topology and RAxML's first two rounds of SPR moves.
Figure 3.
Traversing a tree with up-posteriors.
FastTree optimizes the tree near node N by analyzing the posterior distributions for subtrees A, B, and C, as well as the “up-posterior” D.