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

Table 6 Performance comparison results of AUC values between the combined-features-based models and individual descriptor- and fingerprint-based models

Method	Combined features	AUC	Molecular feature	AUC	Difference
DNN^a	AtomPairs::RDKitDes	0.749	AtomPairs	0.752	− 0.003
	FP2::RDKitDes	0.762	FP2	0.753	0.009
	MACCS::RDKitDes	0.741	MACCS	0.705	0.036
	Morgan::RDKitDes	0.774	Morgan	0.761	0.013
	PharamacoPFP::RDKitDes	0.748	PharamacoPFP	0.735	0.013
			RDKitDes	0.718
KNN^b	AtomPairs::RDKitDes	0.745	AtomPairs	0.743	0.002
	FP2::RDKitDes	0.754	FP2	0.748	0.006
	MACCS::RDKitDes	0.742	MACCS	0.719	0.023
	Morgan::RDKitDes	0.767	Morgan	0.755	0.012
	PharmacoPFP::RDKitDes	0.749	PharmacoPFP	0.740	0.009
			RDKitDes	0.774
NB^c	AtomPairs::RDKitDes	0.738	AtomPairs	0.733	0.005
	FP2::RDKitDes	0.747	FP2	0.743	0.004
	MACCS::RDKitDes	0.750	MACCS	0.724	0.026
	Morgan::RDKitDes	0.781	Morgan	0.772	0.009
	PharmacoPFP::RDKitDes	0.737	PharmacoPFP	0.726	0.011
			RDKitDes	0.763
RF^d	AtomPairs::RDKitDes	0.792	AtomPairs	0.779	0.013
	FP2::RDKitDes	0.803	FP2	0.786	0.017
	MACCS::RDKitDes	0.799	MACCS	0.751	0.048
	Morgan::RDKitDes	0.815	Morgan	0.774	0.041
	PharmacoPFP::RDKitDes	0.801	PharmacoPFP	0.757	0.044
			RDKitDes	0.798
SVM^e	AtomPairs::RDKitDes	0.699	AtomPairs	0.698	0.001
	FP2::RDKitDes	0.686	FP2	0.682	0.004
	MACCS::RDKitDes	0.681	MACCS	0.670	0.011
	Morgan::RDKitDes	0.685	Morgan	0.680	0.005
	PharmacoPFP::RDKitDes	0.687	PharmacoPFP	0.684	0.003
			RDKitDes	0.727
XGBoost^f	AtomPairs::RDKitDes	0.763	AtomPairs	0.759	0.004
	FP2::RDKitDes	0.768	FP2	0.761	0.007
	MACCS::RDKitDes	0.758	MACCS	0.739	0.019
	Morgan::RDKitDes	0.768	Morgan	0.761	0.007
	PharmacoPFP::RDKitDes	0.763	PharmacoPFP	0.748	0.015
			RDKitDes	0.755

ISSN: 1758-2946