A Novel Triple Matrix Factorization Method for Detecting Drug-Side Effect Association Based on Kernel Target Alignment

Table 3

The performance of different kernels via 5-fold Cross-Validation.

Models	Pauwels’s dataset		Mizutani’s dataset		Liu’s dataset
	AUPR	AUC	AUPR	AUC	AUPR	AUC
& ^a	0.4420	0.8950	0.4735	0.9148	0.4718	0.9145
& ^a	0.4892	0.8994	0.5343	0.9070	0.5224	0.9067
& ^a	0.4994	0.8981	0.5217	0.9005	0.5143	0.9026
& ^a	0.4978	0.9079	0.5591	0.9214	0.5529	0.9238
& ^b	0.6254	0.9300	0.6623	0.9376	0.6574	0.9398
& ^b	0.5861	0.9035	0.6324	0.9090	0.6252	0.9087
& ^b	0.5833	0.8999	0.6123	0.9014	0.6047	0.9013
& ^b	0.6557	0.9428	0.6615	0.9369	0.6587	0.9408
Mean weighted^c	0.6598	0.9353	0.6724	0.9280	0.6651	0.9285
KTA-MKL^c	0.6765	0.9434	0.6847	0.9409	0.6801	0.9426

^aThe TMF uses the drug fingerprint and drug profile for side effects. ^bThe TMF uses the side effect profile for drugs and drug profile for side effects. ^cThe TMF uses the drug fingerprint, side effect profile for drugs, and drug profile for side effects.