Syn-MolOpt: a synthesis planning-driven molecular optimization method using data-derived functional reaction templates

Table 2 Overall comparison on multi-property optimization of GSK3β-Mutag

Methods	Property	Top-1^a	Top-10	Top-100	Top-128	Syn-ratio^b ↑
Modof	GSK3β ↑	0.870	0.852 ± 0.047	0.762 ± 0.134	0.736 ± 0.138	0.500
Modof	Mutag ↓	0.057	0.080 ± 0.069	0.445 ± 0.376	0.527 ± 0.387	0.500
HierG2G	GSK3β ↑	0.830	0.872 ± 0.105	0.498 ± 0.277	0.474 ± 0.286	0.651
HierG2G	Mutag ↓	0.092	0.396 ± 0.268	0.417 ± 0.321	0.412 ± 0.319	0.651
SynNet	GSK3β ↑	0.860	0.838 ± 0.022	0.787 ± 0.036	0.780 ± 0.038	1.000 (0.695)^c
SynNet	Mutag ↓	0.080	0.093 ± 0.045	0.134 ± 0.068	0.142 ± 0.072	1.000 (0.695)^c
Syn-MolOpt	GSK3β ↑	0.900	0.899 ± 0.017	0.871 ± 0.021	0.867 ± 0.024	1.000 (0.703)
Syn-MolOpt	Mutag ↓	0.020	0.094 ± 0.009	0.121 ± 0.046	0.128 ± 0.051	1.000 (0.703)

Bold values indicate the highest performance metrics
^aTop-n is the average score for the top n output molecules
^bSyn-ratio is synthesizable ratio
^cThe value in parentheses represents the synthesizable ratio evaluated by AiZynthFinder

ISSN: 1758-2946