Table 1:
Single and multi objective optimization performance comparison on the penalized logP task. For the single one, MolDQN results are from [44], and the scores of other baselines are from [21]. The reported scores of the multi objective optimization task is a success rate. CMG outperforms the baselines in both of the two tasks.
| Single Obj. Opt. | Multi Obj. Opt. | |||
|---|---|---|---|---|
| Method | Improvement | Diversity | All Samples | Sub Samples |
| MMPA | 3.29 ± 1.12 | 0.496 | - | - |
| JT-VAE | 1.03 ± 1.39 | - | - | - |
| GCPN | 2.49 ± 1.30 | - | - | - |
| VSeq2Seq | 3.37 ± 1.75 | 0.471 | - | - |
| MolDQN | 3.37 ± 1.62 | - | - | 0.00% |
| VJTNN | 3.55 ± 1.67 | 0.480 | 3.56% | 4.00% |
| CMG | 3.92 ± 1.88 | 0.545 | 6.98% | 6.00% |