Table 4. Change in the Test Set Performance as the Low Solubility Molecules are Filtered out of the Data Seta.
MDM |
GNN |
|||||||
---|---|---|---|---|---|---|---|---|
log S thresh | R2 | RMSE | Spearman | MAE | R2 | RMSE | Spearman | MAE |
–10 | 0.7654 | 1.0604 | 0.8763 | 0.6975 | 0.7627 | 1.0663 | 0.8740 | 0.7333 |
–7 | 0.7461 | 0.9580 | 0.8644 | 0.6351 | 0.7289 | 0.9899 | 0.8576 | 0.6846 |
–5 | 0.7093 | 0.8423 | 0.8343 | 0.5790 | 0.6910 | 0.8684 | 0.8244 | 0.6153 |
–4 | 0.6817 | 0.7597 | 0.8106 | 0.5447 | 0.6620 | 0.7828 | 0.7955 | 0.5759 |
–2 | 0.5206 | 0.6134 | 0.6850 | 0.4564 | 0.5101 | 0.6201 | 0.6572 | 0.4707 |
Results given in each row were obtained using molecules with log S > “log S thresh”.