Table 2.
Best results per problem. Zero-R classifier results are shown in brackets. The lower part of the Table shows the MCC results from Leong et al.7, for the problem of predicting pathogenicity directly.
| Outcome | Accuracy | AUC | MCC | Method |
|---|---|---|---|---|
| Changed/Unchanged | 66.7% (72.8%) | 0.693 (0.5) | 0.239 (0) | Naive Bayes |
| Activation | 66.1% (62.9%) | 0.657 (0.5) | 0.325 (0) | Naive Bayes |
| Inactivation | 66.7% (60.9%) | 0.730 (0.5) | 0.305 (0) | Naive Bayes |
| Late component | 60.0% (65.0%) | 0.530 (0.5) | 0.061 (0) | k-Nearest neighbour |
| Zero current | 91.4% (87.7%) | 0.785 (0.5) | 0.584 (0) | Naive Bayes |
| Pathogenicity | 0.00 (0) | PolyPhen-2 | ||
| Pathogenicity | 0.00 (0) | SNPs&GO | ||
| Pathogenicity | 0.15 (0) | SIFT | ||
| Pathogenicity | 0.21 (0) | PROVEAN | ||
| Pathogenicity | 0.24 (0) | SNAP | ||
| Pathogenicity | 0.32 (0) | SNAP & Provean |