Table 3.
Performance of ensemble models and some existing software in the external validation dataset.
| Models | Type | Q (%) | SE (%) | SP (%) | AUC (%) |
|---|---|---|---|---|---|
| Ensemble SVM | machine learning | 67.5 | 60.9 | 76.5 | 81.8 |
| Ensemble RF | machine learning | 65.0 | 56.5 | 76.5 | 80.1 |
| Ensemble XGBoost | machine learning | 70.0 | 65.2 | 76.5 | 80.3 |
| admetSAR | machine learning | 50.0 | 34.8 | 70.6 | 49.6 |
| PreADMET | machine learning | 62.5 | 52.2 | 76.5 | —a |
| VEGA CAESAR | machine learning | 70.0 | 65.2 | 76.5 | —a |
| VEGA ISS | rule based | 70.0 | 73.9 | 64.7 | —a |
| VEGA IRFMN/Antares | rule based | 70.0 | 78.3 | 58.8 | —a |
| VEGA IRFMN/ISSCAN-CGX | rule based | 75.0 | 82.6 | 64.7 | —a |
| Toxtree | rule based | 70.0 | 78.3 | 58.6 | —a |
| lazar | similarity search | 75.0 | 87.0 | 58.8 | —a |
aThe AUC cannot be calculated for this software because there are no probability values in its results.