SMILES-based deep generative scaffold decorator for de-novo drug design

Table 1 Results from the decoration process of the DRD2 molecules on the five validation set scaffolds

S	Generated		ChEMBL decoys			DRD2 decoys
S	Total	% act.	% act.	% diff	EOR	% act.	% diff	EOR
1	1728	64.5	0.9	63.6	67.7	4.2	60.3	15.5
2	1626	52.4	8.3	44.1	6.3	19.7	32.7	2.7
3	10,666	35.6	0.8	34.8	46.3	16.2	19.4	2.2
4	262	84.7	0.4	84.3	236.1	0.9	83.8	89.0
5	18	88.9	7.8	81.1	11.3	10.1	78.8	8.8

Total number of molecules sampled (Total); percent of generated molecules that are predicted as active (\(p_{active} \ge 0.5\)) by the APM (% act); for both the decoys decorated with ChEMBL fragments and DRD2 fragments from the training set: percent of predicted active decoys (\(p_{active} \ge 0.5\)) (% act); difference between the generated predicted active percent and the predicted active percent of the decoys (% Diff); Enrichment Over Random (\(percent_{active} /percent_{decoy}\)) (EOR)

ISSN: 1758-2946