Table 1 Comparison of model performance and properties for non-sulphur containing structures generated by the two models
From: Molecular de-novo design through deep reinforcement learning
Model | Prior | Agent | Action basis | REINFORCE | REINFORCE + Prior |
|---|---|---|---|---|---|
Fraction of valid SMILES | 0.94 ± 0.01 | 0.95 ± 0.01 | 0.95 ± 0.01 | 0.98 ± 0.00 | 0.98 ± 0.00 |
Fraction without sulphur | 0.66 ± 0.01 | 0.98 ± 0.00 | 0.92 ± 0.02 | 0.98 ± 0.00 | 0.92 ± 0.01 |
Average molecular weight | 371 ± 1.70 | 367 ± 3.30 | 372 ± 0.94 | 585 ± 27.4 | 232 ± 5.25 |
Average cLogP | 3.36 ± 0.04 | 3.37 ± 0.09 | 3.39 ± 0.02 | 11.3 ± 0.85 | 3.05 ± 0.02 |
Average NumRotBonds | 5.39 ± 0.04 | 5.41 ± 0.07 | 6.08 ± 0.04 | 30.0 ± 2.17 | 2.8 ± 0.11 |
Average NumAromRings | 2.26 ± 0.02 | 2.26 ± 0.02 | 2.09 ± 0.02 | 0.57 ± 0.04 | 2.11 ± 0.02 |