Table 4.
Comparison with external predictors in the test set built by Veltri et al. [30]. Redundant instances with our training set were removed. The values in bold denote the best performance for a given measure.
| Webserver | Algorithm | Pr. | Sn. | Acc. |
|---|---|---|---|---|
| ClassAMP | SVM | 0.36 | 0.27 | 0.66 |
| MLAMP | RF | 0.51 | 0.65 | 0.72 |
| iAMPred | SVM | 0.74 | 0.90 | 0.88 |
| AMPScanner vr.1 # | RF | 0.64 | 0.77 | 0.81 |
| AMPScanner vr.2 * | DNN | 0.82 | 0.89 | 0.91 |
| AMPDiscover | RF | 0.83 | 0.84 | 0.91 |
| ABP-Finder (Training) | RF | 0.83 | 0.43 | 0.81 |
| ABP-Finder (Training, AD) | RF | 0.83 | 0.51 | 0.86 |
| ABP-Finder (Production) | RF | 0.84 | 0.48 | 0.83 |
| ABP-Finder (Production, AD) | RF | 0.84 | 0.57 | 0.86 |
AD: only instances within our applicability domain are considered valid predictions. # AMPScanner_v1 only considers peptides ≥ 10 AA for the prediction. * Performance based on the model from the original training in Veltri et al. [30], where the cases in this test set are held out of the training process.