CardioTox net: a robust predictor for hERG channel blockade based on deep learning meta-feature ensembles

Table 4 Comparison of CardioTox with other methods using three external independent test sets. B-ACC refers to balanced accuracy

Evaluation data	Methods	MCC	NPV	ACC	PPV	SPE	SEN	B-ACC
Test set-I	CardioTox	0.599	0.688	0.810	0.893	0.786	0.833	0.810
	DeepHIT	0.476	0.643	0.773	0.833	0.643	0.833	0.738
	CardPred	0.193	0.421	0.614	0.760	0.571	0.633	0.602
	OCHEM Predictor-I	0.149	0.333	0.364	1.000	1.000	0.067	0.534
	OCHEM Predictor-II	0.164	0.351	0.432	0.857	0.929	0.200	0.564
	Pred-hERG 4.2	0.306	0.538	0.705	0.774	0.500	0.800	0.650
	Random Forest	0.092	0.375	0.547	0.714	0.428	0.666	0.547
	Support Vector Machines	0.00	0.318	0.5	NA	1.0	0.0	0.500
	Gradient Boosting	0.133	0.375	0.571	0.750	0.642	0.500	0.571
Test set-II	CardioTox	0.452	0.947	0.755	0.455	0.600	0.909	0.754
	DeepHIT	0.398	0.941	0.721	0.417	0.533	0.909	0.721
	CardPred	0.049	0.750	0.527	0.294	0.600	0.454	0.527
	OCHEM Predictor-I	0.372	0.800	0.648	0.666	0.933	0.364	0.648
	OCHEM Predictor-II	0.310	0.794	0.632	0.571	0.900	0.364	0.632
	Pred-hERG 4.2	0.146	0.813	0.580	0.320	0.433	0.727	0.580
	Random Forest	0.397	0.941	0.721	0.416	0.533	0.909	0.721
	Support Vector Machines	0.345	0.933	0.683	0.384	0.466	0.909	0.687
	Gradient Boosting	0.421	0.941	0.727	0.423	0.550	0.909	0.729
Test set-III	CardioTox	0.220	0.986	0.746	0.113	0.698	0.794	0.746
	DeepHIT	0.134	0.981	0.660	0.074	0.526	0.794	0.660
	CardPred	0.112	0.975	0.633	0.072	0.561	0.705	0.633
	OCHEM Predictor-I	0.162	0.959	0.562	0.25	0.978	0.147	0.562
	OCHEM Predictor-II	0.108	0.961	0.575	0.117	0.915	0.235	0.575
	Pred-hERG 4.2	0.151	0.986	0.680	0.077	0.508	0.852	0.680
	Random Forest	0.179	0.985	0.714	0.085	0.546	0.882	0.714
	Support Vector Machines	0.147	0.981	0.668	0.068	0.365	0.970	0.667
	Gradient Boosting	0.182	0.981	0.701	0.083	0.551	0.882	0.716

ISSN: 1758-2946