TABLE 1.
Antibody affinity classifier ccross-validation performance across feature-sets. Summary of XGBoost (XGB) classifiers trained on each of the feature-sets, including feature numbers and classifier cross-validation (AUC and F-1 score). Performance of different classifier architectures, including RandomForest (RF), Support Vector Classifier (SVC), and Multi-layer Perceptron (MLP) are also shown for the "combined" feature-sets.
| Feature-set | #Features in set (#used) | AUC/F1 |
|---|---|---|
| Energetics | 18 (10) | 0.62/0.67 |
| dMaSIF-site | 26 (10) | 0.62/0.67 |
| Network (SIN) | 26 (10) | 0.69/0.70 |
| Statistical (AIF) | 26 (10) | 0.58/0.67 |
| aa_counts | 400 (10) | 0.61/0.64 |
| aa_counts_CDR | 150 (10) | 0.67/0.67 |
| num_multivalent | 7 (7) | 0.57/0.64 |
| Ab_info | 47 (10) | 0.60/0.63 |
| Combined-XGB | 700 (16) | 0.72/0.72 |
| Combined-RF | 0.70/0.73 | |
| Combined-SVC | 0.66/0.74 | |
| Combined-MLP | 0.67/0.71 |