Table 3. The results from the two times five-fold DTU Food cross-validation procedure of the cocktail models with different active-to-inactive ratios.
| Training set | QSAR2:1 | QSAR3:1 | QSAR4:1 | QSAR4:1-R | Rational-final | Random-final | |
|---|---|---|---|---|---|---|---|
| Predictions in LPDM and DTU Food domain | TP | 982 | 724 | 648 | 994 | 775 | 1,132 |
| TN | 2,247 | 2,58 | 3,079 | 4,657 | 3,679 | 5,228 | |
| FP | 263 | 423 | 513 | 432 | 552 | 486 | |
| FN | 204 | 248 | 281 | 188 | 264 | 178 | |
| AD, % | 74.0±1.7 | 59.7±2.4 | 54.3±2.1 | 75.4±1.2 | 57.0±1.4 | 75.9±1.1 | |
| Cooper statistics | Sens., % | 82.8±3.4 | 74.5±4.3 | 69.6±5.5 | 84.0±3.8 | 74.5±4.7 | 86.4±2.6 |
| Spec., % | 89.5±1.8 | 86.0±2.9 | 85.7±2.5 | 91.5±1.8 | 87.0±1.6 | 91.5±0.8 | |
| BA, % | 86.1±2.0 | 80.2±2.3 | 77.6±3.0 | 87.8±2.2 | 80.8±2.5 | 88.9±1.0 | |
| Prevalence of actives 1% | PPV, % | 7.4 | 5.1 | 4.7 | 9.1 | 5.5 | 9.3 |
| NPV, % | 99.8 | 99.7 | 99.6 | 99.8 | 99.7 | 99.9 | |
| Prevalence of actives 5% | PPV, % | 29.3 | 21.9 | 20.4 | 34.2 | 23.2 | 34.9 |
| NPV, % | 99.0 | 98.5 | 98.2 | 99.1 | 98.5 | 99.2 | |
| Prevalence of actives 10% | PPV, % | 46.7 | 37.2 | 35.1 | 52.3 | 38.9 | 53.0 |
| NPV, % | 97.9 | 96.8 | 96.2 | 98.1 | 96.8 | 98.4 |
TP = true positive, TN = true negative, Ttotal = total number of true predictions, FP = false positive, FN = false negative, Ftotal = total number of false predictions, Sens. = sensitivity, Spec. = specificity, BA = balanced accuracy, PPV = positive predictive value (the percentage of true positives among the predicted positives), NPV = negative predictive value (the percentage of true negatives among the predicted negatives)