Table 3.

Statistical parameters for the consensus models

Modela	Dataset	TPb	FPc	TNd	FNe	TP + TN	Totalf	Qg	Sens.h	Spec.i	Prec.j	G-meank
1	Training	72	25	54	21	126	172	0.73	0.77	0.68	0.74	0.73
	Validation	130	654	920	91	1050	1795	0.58	0.59	0.58	0.17	0.59
2	Training	73	31	48	20	121	172	0.70	0.78	0.61	0.70	0.69
	Validation	140	723	851	81	991	1795	0.55	0.63	0.54	0.16	0.59
3	Training	71	31	48	22	119	172	0.69	0.76	0.61	0.70	0.68
	Validation	135	707	867	86	1002	1795	0.56	0.61	0.55	0.16	0.58
4	Training	74	32	47	19	121	172	0.70	0.80	0.59	0.70	0.69
	Validation	128	718	856	93	984	1795	0.55	0.58	0.54	0.15	0.56
5	Training	73	29	50	20	123	172	0.72	0.78	0.63	0.72	0.70
	Validation	132	685	889	89	1021	1795	0.57	0.60	0.56	0.16	0.58
6	Training	73	28	51	20	124	172	0.72	0.78	0.65	0.72	0.71
	Validation	131	675	899	90	1030	1795	0.57	0.59	0.57	0.16	0.58

^aModel 1 = substructure (SS) + substructure count (SSC) + extended CDK (ECDK), 2 = PubChem (PC) + SSC + ECDK, 3 = PC + SSC + SS, 4 = PC + SSC + MACCS, 5 = PC + SSC + ECDK + SC + MACCS, 6 = PC + SSC + ECDK + SS + MACCS + CDK + CDK Graph, ^b true positives, ^c false positives, ^d true negatives, ^e false negatives, ^f TP + TN + FP + FN, ^g overall accuracy of prediction, ^h sensitivity, ⁱ specificity, ^j precision, ^k  $\sqrt{\text{Sensitivity}\times \text{Specificity}}$