Table 2.
Comparison of algorithms tested on Swiss-Prot dataset*
| AUC | Accuracy | Precision | Sensitivity | Specificity | |
|---|---|---|---|---|---|
| EFIN (Swiss-Prot)** | 90.7%(1.0%) | 83.7%(0.98%) | 86.7%(3.3%) | 86.4%(0.9%) | 79.5%(3.0%) |
| GERP++ | 76.10% | 52.80% | 45.27% | 96.78% | 24.36% |
| PhyloP | 76.30% | 54.47% | 46.18% | 96.47% | 27.33% |
| MutationTaster | 85.40% | 79.47% | 69.07% | 86.42% | 74.98% |
| SIFT | 83.60% | 76.58% | 67.29% | 78.52% | 75.32% |
*This test is based on a subset of Swiss-Prot, of which mutations can be processed by all these 5 tools, including totally 18660 damaging variants and 28863 neutral variants.
**Swiss-Prot trained EFIN is validated by 10 fold cross-validation, and all these statistical measures are average values. Standard deviation is described within brackets after each measure.