Large-scale comparison of machine learning methods for profiling prediction of kinase inhibitors

Table 3 Performance comparison results of RDKit descriptor-based predictive models on the test sets of 354 kinases

Molecular feature	Method	AUC^g	F1 score^h	BAⁱ
RDKitDes	RF^a	0.798 ± 0.120	0.759 ± 0.225	0.650 ± 0.113
	NB^b	0.763 ± 0.099	0.739 ± 0.155	0.681 ± 0.090
	SVM^c	0.727 ± 0.206	0.723 ± 0.245	0.611 ± 0.165
	KNN^d	0.774 ± 0.116	0.776 ± 0.186	0.684 ± 0.104
	XGBoost^e	0.755 ± 0.148	0.747 ± 0.216	0.650 ± 0.117
	DNN^f	0.718 ± 0.180	0.693 ± 0.254	0.589 ± 0.117
	Mean	0.756 ± 0.030	0.740 ± 0.029	0.644 ± 0.038

ISSN: 1758-2946