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

Table 3 Descriptors calculated for the molecules decorated with both model architectures (single and multi-step) from the validation set scaffolds and the non-dataset scaffolds

Set	Mols/scaff.	A (%)	B (%)	C (%)	D (%)	E (%)
Multi-step decorator model
Validation set scaff.	12,294	97.9	82.7	99.6	68.3	98.2
Non-dataset scaff.	11,504	99.2	89.4	99.8	78.8	98.7
Single-step decorator model
Validation set scaff.	38,344	95.9	63.2	99.6	52.7	98.5
Non-dataset scaff.	25,462	97.9	66.1	99.8	57.3	98.7

Molecules per scaffold (Mols/scaff.); See the list above for information on the other fields
A Percent of decorated scaffolds with all attachment point bonds RECAP compliant
B Percent of decorated scaffolds with all decorations in the training set
C Percent of decorated scaffolds with at most one decoration not in the training set
D Percent of decorated scaffolds with all decorations in ZINC in-stock
E Percent of decorated scaffolds with at most one decoration not in ZINC in-stock

ISSN: 1758-2946