PhyML 3.0 Benchmarks
aLRT versus bootstrap branch supports
Comparison of alRT-based and bootstrap branch supports, using 100 DNA and protein alignments.
Bootstrap and aLRT-SH agreement as a function of the phylogenetic signal: the phylogenetic signal is measured by the number of sites (with less than 10% gaps or missing values) times the median of internal branch lengths. This roughly corresponds to the expected number of substitutions supporting any given internal branch. Branch support agreement equals the proportion of branches with both SH-like support > 0.90 and bootstrap support > 0.75.
Data sets
The benchmark contains 50 protein alignments and 50 DNA alignments.
- DNA alignments
We selected the 50 most recent alignments from Treebase with at least 50 sequences, less than 200 sequences and less than 2000 sites.
- Protein alignments
We selected the 50 most recent alignments from Treebase with at least 5 sequences, less than 200 sequences and less than 2000 sites.
Programs
- PhyML 3.0
PhyML 3.0 is used with: SPR (and NNI) topological moves; 100 bootstrap replicates and aLRT SH-like supports; LG model for proteins and GTR for DNA sequences; 4 discrete gamma rate categories with alpha equal to 0.5.
Results
- Graphics
For each alignment, we provide a graphic showing the bootstrap support as a function of the aLRT statistic. Dots are colored depending on the SH-like support, and vertical lines indicate the chi-square-based support.
- Phylogenetic signal/ Bootstrap and aLRT-SH agreement
The agreement between bootstrap and SH-like supports depends on the phylogenetic signal. This is measured by the product of the number of sites in the alignment, times the median value of the internal branch lengths. Only sites with less than 10% gaps are accounted for. The support agreement is measured using the proportion of branches with both bootstrap > 0.75 and SH-like support > 0.90. The higher the signal, the higher is the agreement (see figure).
- RAxML/PhyML bootstrap support distributions
For all data sets with median value of internal branch lengths equal to 0, and for the two data sets of figure 6 (M1499 and M2588), we display graphics showing the distributions of RAxML vs PhyML bootstrap values. We have used fast RAxML Blackbox server to compute RAxML bootstrap values. These graphics clearly show that RAxML tends to support very short (zero-length) branches even more than PhyML does. This is likely explained by hidden determinisms which we tried to eliminate in PhyML as much as possible. With high signal and long branches (M1499), RAxML and PhyML bootstrap values seems to be quite congruent.