Table 4.

Statistics of the SVM-model and commercial models for the test set.

Name	TP/FN FP/TN	Accuracy	Sensitivity	Specificity	BAC	Kappa	ROC
SVM-72 model*	1,987/979 410/83,985	0.984	0.670	0.995	0.833	0.733	0.962
	1,838/854 327/74,043	0.985	0.683	0.996	0.839	0.749	0.966
ACD/Percepta	1,890/802 6,518/67,852	0.905	0.702	0.912	0.807	0.304	0.890
ADMET Predictor	2,329/363 25,495/48,875	0.664	0.865	0.657	0.761	0.095	0.866
StarDrop	2,357/335 32,963/41,407	0.568	0.876	0.557	0.716	0.063	0.831

^*For the SVM model, the first row represents the results for all test set compounds, and the second row represents the results for 77,062 compounds used with all three commercial software programs to predict the hERG inhibitory activity.