Table 5 Ablation studies
From: Force field-inspired molecular representation learning for property prediction
Method | Regression tasks (RMSE) (RMSE, lower is better) | Classification tasks (ROC-AUC, higher is better) | ||||
|---|---|---|---|---|---|---|
ESOL | Lipophilicity | FreeSolv | PDBBind | BACE | BBBP | |
FFiNet-no-axial | 0.638 (0.048) | 0.603 (0.037) | 1.019 (0.283) | 1.624 (0.045) | 0.869(0.010) | 0.847 (0.021) |
FFiNet-1hop | 0.614 (0.047) | 0.685 (0.088) | 0.951 (0.010) | 1.437 (0.031) | 0.876 (0.024) | 0.897 (0.015) |
FFiNet-2hop | 0.607 (0.039) | 0.648 (0.076) | 0.808 (0.148) | 1.392 (0.044) | 0.856 (0.022) | 0.907 (0.019) |
FFiNet | 0.551 (0.030) | 0.579 (0.022) | 0.756 (0.138) | 1.310 (0.012) | 0.891 (0.016) | 0.916 (0.012) |