The influence of the experts' prior knowledge on the search resultsa.
| PDB | Ligand score (kcal mol−1) | Number of molecules (↑) | Average score (kcal mol−1, ↓) | QED (↑) | SA (↑) | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| Test set | Baseline (Pocket2Mol) | Off | On | Off | On | Off | On | Off | On | |
| 1fmc | −9.32 | −6.92 | 88 | 160 | −9.52 | −9.94 | 0.69 | 0.63 | 0.80 | 0.74 |
| 3g51 | −7.63 | −8.31 | 135 | 167 | −9.55 | −9.90 | 0.65 | 0.63 | 0.79 | 0.76 |
| 5q0k | −9.24 | −6.67 | 15 | 58 | −9.52 | −9.90 | 0.63 | 0.66 | 0.80 | 0.80 |
| 1phk | −8.14 | −10.72 | 433 | 737 | −9.52 | −10.09 | 0.65 | 0.66 | 0.76 | 0.75 |
| 4f1m | −6.58 | −6.78 | 18 | 46 | −9.29 | −9.78 | 0.70 | 0.58 | 0.79 | 0.79 |
| 1e8h | −8.83 | −8.72 | 264 | 353 | −9.55 | −10.08 | 0.70 | 0.65 | 0.78 | 0.77 |
| 1ai4 | −6.28 | −5.60 | 114 | 81 | −9.41 | −9.86 | 0.61 | 0.63 | 0.82 | 0.81 |
| 3w83 | −3.91 | −3.30 | 51 | 83 | −9.54 | −9.86 | 0.64 | 0.60 | 0.78 | 0.74 |
| 5d7n | −7.56 | −7.71 | 128 | 114 | −9.43 | −9.97 | 0.63 | 0.65 | 0.78 | 0.77 |
| 5aeh | −11.93 | −10.11 | 593 | 843 | −10.26 | −10.78 | 0.64 | 0.68 | 0.73 | 0.79 |
| 4qlk | −8.48 | −7.34 | 238 | 366 | −9.48 | −9.97 | 0.57 | 0.66 | 0.76 | 0.80 |
| 3hy9 | −8.12 | −6.66 | 528 | 768 | −9.51 | −10.06 | 0.64 | 0.65 | 0.79 | 0.80 |
| 4m7t | −7.81 | −8.24 | 525 | 701 | −9.69 | −10.07 | 0.65 | 0.64 | 0.76 | 0.78 |
| Average | −7.99 | −7.47 | 241 | 344 | −9.56 | −10.02 | 0.65 | 0.64 | 0.78 | 0.78 |
Off represents the standard 3D-MCTS, which only considers binding affinity as the reward; On represents a version of 3D-MCTS that incorporates domain knowledge based on pharmacophores during the generation process.