Table 8.
Results for tree inference on alignments generated using different methods, on the simulated datasets, as shown in Figure 15
| Low | Medium | High | |||||
|---|---|---|---|---|---|---|---|
| AMA | RF | AMA | RF | AMA | RF | ||
| MinRisk (C +), g=0 | 0.84 | 0.12 | 0.67 | 0.12 | 0.59 | 0.44 | |
| MinRisk (C), g=0 | 0.83 | 0.20 | 0.64 | 0.28 | 0.53 | 0.44 | |
| MAFFT | 0.81 | 0.16 | 0.60 | 0.44 | 0.50 | 0.64 | |
| MUSCLE | 0.80 | 0.36 | 0.61 | 0.48 | 0.49 | 0.48 | |
| T-Coffee | 0.67 | 1.00 | 0.52 | 0.92 | 0.42 | 1.24 | |
| CLUSTALW2 | 0.62 | 1.04 | 0.44 | 1.32 | 0.35 | 1.36 | |
Shown are alignment accuracy scores (according to the AMA metric), and Robinson-Foulds tree distances (RF) for the DNAML tree, averaged over all datasets in each group (low, medium and high indel rates). Higher alignment accuracy is strongly predictive of tree accuracy, with the most accurate alignments generating the trees with the fewest errors (shown in bold). The MinRisk results were computed using samples generated by StatAlign.