Performance of AIxFuse and compared methods on RORγt|DHODH dual-target drug design benchmark.
| Metrics | REINVENT2.0 | RationaleRL | MARS | AIxFuse |
|---|---|---|---|---|
| Validity (%) | 100 | 99.9 | 100 | 100 |
| Uniqueness (%) | 33.5 | 33.7 | 9.9 | 94.1 |
| Diversity | 0.708 | 0.672 | 0.530 | 0.661 |
| Success rate (%) | 4.65 | 1.16 | 0.00 | 23.96 |
| USR QED (%) | 25.94 | 1.39 | 9.83 | 41.53 |
| USR SA (%) | 31.39 | 33.58 | 9.77 | 82.41 |
| USR dockingRORγt (%) | 21.60 | 23.56 | 1.67 | 85.04 |
| USR dockingDHODH (%) | 4.94 | 16.54 | 0.02 | 72.74 |
| USR 3D SNNRORγta (%) | 20.33 | 12.83 | 3.17 | 58.18 |
| USR 3D SNNDHODH (%) | 22.81 | 10.71 | 6.95 | 83.87 |
a
3D SNN measures the maximum 3D similarity to known active compounds, while USR 3D SNN is the proportion of non-repeating compounds that obtained higher 3D SNN than dual-target inhibitor (R)-14d.3