Table 4.

Augmentation effect on architecture C biLSTM–biLSTM with layer sizes 64/64 and 4 concatenated encoding layers

Smiles	Augm.	Best model epoch#	Validity%	Uniqueness%	Training%	Length match%a	HAC match%b
Canonical	1	9, 9, 7	96.6 ± 0.5	99.9 ± 0.1	16.2 ± 1.5	93.3 ± 0.3	92.0 ± 0.5
Random	1	10, 14, 16	97.0 ± 0.3	99.9 ± 0.0	11.9 ± 0.6	98.5 ± 0.3	97.4 ± 0.5
Random	2	5, 5, 5	97.3 ± 0.1	99.9 ± 0.0	13.9 ± 0.5	97.7 ± 0.4	94.5 ± 0.8
Random	3	4, 6, 4	97.9 ± 0.3	99.9 ± 0.0	13.6 ± 0.5	98.8 ± 0.1	96.5 ± 0.2
Random	4	4, 3, 4	98.2 ± 0.4	99.9 ± 0.0	11.6 ± 0.5	98.8 ± 0.3	97.1 ± 0.2
Random	5	4, 4, 4	98.3 ± 0.3	99.9 ± 0.0	11.2 ± 0.5	97.3 ± 0.7	96.6 ± 0.3
Random	10	4, 4, 4	98.3 ± 0.3	99.9 ± 0.0	14.2 ± 0.5	98.4 ± 0.4	98.2 ± 0.5

^aLength match for SMILES length distributions of the training set and generated set (See “Methods”)

^bHAC match for the atom count distributions of the generated set and training set (See “Methods”)