Table 1.
Performance of 14 binary classifiers for predicting hemolytic activity using 3 model datasets (HemoPI-1, HemoPI-2, HemoPI-3) and 56 sequence-based physicochemical descriptors.
| Classifiers | Mean accuracy (%) | |||||
|---|---|---|---|---|---|---|
| HemoPI-1 dataset | HemoPI-2 dataset | HemoPI-3 dataset | ||||
| Model | Validation | Model | Validation | Model | Validation | |
| LOGREG | 92.6 | 89.6 | 68.0 | 65.8 | 69.7 | 70.8 |
| KNN | 93.2 | 86.8 | 71.6 | 63.4 | 72.8 | 65.8 |
| CART | 90.4 | 83.7 | 69.6 | 62.4 | 68.9 | 64.0 |
| RFC | 93.8 | 87.3 | 69.0 | 63.9 | 72.4 | 66.8 |
| GBC | 94.0 | 90.4 | 76.7 | 72.3 | 75.8 | 72.9 |
| ABC | 93.8 | 91.4 | 71.3 | 73.8 | 72.3 | 71.4 |
| LDA | 94.2 | 90.5 | 70.0 | 59.9 | 70.8 | 70.8 |
| QDA | 91.5 | 85.5 | 66.4 | 60.9 | 65.9 | 59.4 |
| NB | 85.9 | 85.5 | 62.3 | 63.4 | 64.4 | 65.9 |
| SVC-LIN | 88.1 | 85.0 | 61.9 | 59.9 | 65.4 | 66.5 |
| SVC-RBF | 88.7 | 85.5 | 63.7 | 62.4 | 66.3 | 66.2 |
| SVC-POLY | 71.5 | 60.0 | 54.4 | 54.5 | 54.6 | 54.5 |
| SVC-SIG | 88.0 | 81.8 | 59.6 | 54.5 | 63.4 | 63.4 |
| XGBC | 94.8 | 92.4 | 76.1 | 69.3 | 74.7 | 73.2 |
All performance metrics (Accuracy, Precision, Matthews correlation coefficient, Cohen’s kappa statistic and Receiver operating characteristic area under curve) are available in Supporting Information.
The best binary classifiers and their respective performances are depicted in bold.