Table 2.
Multi property molecule generation, tested with the GuacaMol dataset.
| Model | Valid ↑ | Unique ↑ | Novelty ↑ | SD/MAD ↓ | |||
|---|---|---|---|---|---|---|---|
| logP | SAS | TPSA | |||||
| logP + SAS | MolGPT | 0.972 | 0.992 | 1 | 0.340/0.250 | 0.210/0.140 | |
| ours | 0.971 | 0.991 | 1 | 0.331/0.252 | 0.208/0.151 | ||
| SAS + TPSA | MolGPT | 0.971 | 0.988 | 1 | 0.220/0.150 | 4.940/3.760 | |
| ours | 0.970 | 0.988 | 1 | 0.217/0.158 | 4.705/3.647 | ||
| logP + TPSA | MolGPT | 0.965 | 0.994 | 1 | 0.320/0.240 | 4.770/3.710 | |
| ours | 0.961 | 0.995 | 1 | 0.314/0.240 | 4.575/3.607 | ||
| logP + TPSA + SAS | MolGPT | 0.973 | 0.969 | 1 | 0.350/0.270 | 0.260/0.180 | 4.800/3.790 |
| ours | 0.966 | 0.964 | 1 | 0.335/0.259 | 0.247/0.183 | 4.461/3.532 | |