DeepSA: a deep-learning driven predictor of compound synthesis accessibility

Table 2 Performance comparison of the different models on the external three test sets

Datasets	Model	ACC	Recall	Precision	F-score	AUROC	Threshold
TS1	DeepSA	0.995	1.000	0.989	0.995	1.000	0.47
	DeepSA	0.995	1.000	0.990	0.995	1.000	0.50
	GASA	0.987	0.999	0.976	0.987	1.000	0.50
	SAscore	0.989	0.992	0.986	0.989	0.999	4.50
	SAscore	0.665	0.331	0.998	0.497	0.999	6.00
	RAscore	0.919	0.867	0.967	0.914	0.982	0.50
	SYBA	0.962	1.000	0.930	0.964	0.998	0.00
	SCScore	0.608	0.698	0.592	0.641	0.641	3.10
TS2	DeepSA	0.840	0.746	0.861	0.799	0.913	0.47
	DeepSA	0.838	0.730	0.871	0.795	0.913	0.50
	GASA	0.796	0.677	0.815	0.740	0.876	0.50
	SAscore	0.815	0.603	0.946	0.737	0.919	3.40
	SAscore	0.664	0.216	0.996	0.355	0.919	6.00
	RAscore	0.751	0.485	0.878	0.625	0.865	0.50
	SYBA	0.787	0.627	0.834	0.716	0.862	0.00
	SCScore	0.395	0.442	0.341	0.385	0.373	2.30
TS3	DeepSA	0.819	0.761	0.861	0.808	0.896	0.47
	DeepSA	0.817	0.753	0.864	0.805	0.896	0.50
	GASA	0.760	0.646	0.837	0.729	0.849	0.50
	SAscore	0.577	0.211	0.788	0.333	0.772	3.10
	SAscore	0.512	0.044	0.690	0.084	0.772	6.00
	RAscore	0.701	0.571	0.772	0.656	0.790	0.50
	SYBA	0.647	0.387	0.806	0.523	0.790	0.00
	SCScore	0.472	0.723	0.481	0.578	0.425	2.20

ISSN: 1758-2946