Table 2.
Benchmarking BarlowDTI against other models using Koh et al. splits [16]
| Dataset | Split | Model | ROC AUC | PR AUC |
|---|---|---|---|---|
| BioSNAP | Unseen protein | BarlowDTI | 0.9572 | 0.9679 |
| DrugBAN [16, 44] | 0.7327 | 0.7971 | ||
| PSICHIC [16] | 0.8819 | 0.9071 | ||
| STAMP-DPI [16, 45] | 0.8372 | 0.8738 | ||
| XGBoost | 0.8506 | 0.8794 | ||
| Random split | BarlowDTI | 0.9718 | 0.9755 | |
| DrugBAN [16, 44] | 0.9089 | 0.9159 | ||
| PSICHIC [16] | 0.9246 | 0.9226 | ||
| STAMP-DPI [16, 45] | 0.8993 | 0.9056 | ||
| XGBoost | 0.9146 | 0.9242 | ||
| Unseen ligand | BarlowDTI | 0.9666 | 0.9706 | |
| DrugBAN [16, 44] | 0.8775 | 0.8843 | ||
| PSICHIC [16] | 0.9019 | 0.9030 | ||
| STAMP-DPI [16, 45] | 0.8902 | 0.8915 | ||
| XGBoost | 0.8909 | 0.9026 | ||
| BindingDB | Unseen protein | BarlowDTI | 0.6939 | 0.5791 |
| DrugBAN [16, 44] | 0.6523 | 0.5295 | ||
| PSICHIC [16] | 0.7537 | 0.6241 | ||
| STAMP-DPI [16, 45] | 0.6828 | 0.5735 | ||
| XGBoost | 0.6460 | 0.5233 | ||
| Random split | BarlowDTI | 0.9640 | 0.9513 | |
| DrugBAN [16, 44] | 0.9640 | 0.9539 | ||
| PSICHIC [16] | 0.9503 | 0.9280 | ||
| STAMP-DPI [16, 45] | 0.9318 | 0.9085 | ||
| XGBoost | 0.9582 | 0.9462 | ||
| Unseen ligand | BarlowDTI | 0.9456 | 0.9263 | |
| DrugBAN [16, 44] | 0.9409 | 0.9188 | ||
| PSICHIC [16] | 0.9264 | 0.8975 | ||
| STAMP-DPI [16, 45] | 0.9027 | 0.8683 | ||
| XGBoost | 0.9374 | 0.9141 | ||
| Human | Unseen protein | BarlowDTI | 0.9630 | 0.9693 |
| DrugBAN [16, 44] | 0.9298 | 0.9417 | ||
| PSICHIC [16] | 0.9503 | 0.9595 | ||
| STAMP-DPI [16, 45] | 0.8563 | 0.8748 | ||
| XGBoost | 0.8961 | 0.9171 | ||
| Random split | BarlowDTI | 0.9917 | 0.9905 | |
| DrugBAN [16, 44] | 0.9841 | 0.9753 | ||
| PSICHIC [16] | 0.9861 | 0.9840 | ||
| STAMP-DPI [16, 45] | 0.9659 | 0.9582 | ||
| XGBoost | 0.9813 | 0.9782 | ||
| Unseen ligand | BarlowDTI | 0.9346 | 0.9348 | |
| DrugBAN [16, 44] | 0.9459 | 0.9387 | ||
| PSICHIC [16] | 0.9500 | 0.9371 | ||
| STAMP-DPI [16, 45] | 0.9156 | 0.8980 | ||
| XGBoost | 0.9391 | 0.9337 |
Performance was evaluated against three established benchmarks, and the mean of the BarlowDTI performance of five replicates are presented. All other metrics are taken from Koh et al. Best result per benchmark and split is highlighted in bold. Koh et al. does not present replicates or sample-correlated predictions [16]