Table 1. Statistical performance of SVM classification models for substrate or inhibitor of pharmacokinetics-relevant protein, P-gp and CYP.

Model	SwissADME					Previous models
	TR/TS[a]	ACCCV[b]	AUCCV[c]	ACCext[d]	AUCext[e]	TR/TS	ACCCV	AUCCV	ACCext	AUCext	Reference
P-gp substrate	1033/415	0.72	0.77	0.89	0.94	544/n.c.	0.71[f]	n.c.	n.c.	n.c.	53
						484/300	0.64	n.c.	0.59	n.c.	71
						332/n.c.	0.74[f]	0.77[f]	n.c.	n.c.	15
						212/120	0.74	n.c.	0.88	n.c.	57
CYP1A2 inhibitor	9145/3000	0.83	0.90	0.84	0.91	9145/3000	n.c.	n.c.	0.88	0.95	54
						12099/2804	0.82[f]	n.c.	0.68	0.81	77
						7208/7128	0.88[g]	n.c.	n.c.	0.93	55
CYP2C19 inhibitor	9272/3000	0.80	0.86	0.80	0.87	9272/3000	n.c.	n.c.	0.85	0.91	54
						11885/2691	0.79[f]	n.c.	0.81	0.84	77
						6038/5923	0.81[g]	n.c.	n.c.	0.89	55
CYP2C9 inhibitor	5940/2075	0.78	0.85	0.71	0.81	8720/3000	n.c.	n.c.	0.83	0.90	54
						12130/2579	0.78[f]	n.c	0.89	0.86	77
						6627/6530	0.83[g]	n.c.	n.c.	0.89	55
CYP2D6 inhibitor	3664/1068	0.79	0.85	0.81	0.87	9726/3000	n.c.	n.c.	0.84	0.88	54
						11881/2860	0.84[f]	n.c.	0.88	0.88	77
						7788/7761	0.90[g]	n.c.	n.c.	0.85	55
CYP3A4 inhibitor	7518/2579	0.77	0.85	0.78	0.86	8893/5135	n.c.	n.c.	0.84	0.92	54
						11536/7025	0.78[f]	n.c.	0.76	0.78	77
						2334/6738	0.81[g]	n.c.	n.c.	0.87	55

^aNumber of molecules in the training set (TR) and in the test set (TS); ^b10-fold cross-validation accuracy; ^c10-fold cross-validation area under receiver operating characteristic (ROC) curve; ^dexternal validation accuracy; ^eexternal validation area under ROC curve; ^f5-fold cross-validation; ^g7-fold cross-validation.