Predicting potential drug-drug interactions by integrating chemical, biological, phenotypic and network data

BMC Bioinformatics

Table 5 Performances of the ensemble method and benchmark methods evaluated by 20 runs of 3-CV and 5-CV

Evluation	Method	AUC	AUPR	Precision	Recall	Accuracy	F-measure
3-CV evaluation	Vilar’s substructure-based model	0.670	0.273	0.145	0.535	0.684	0.229
	Vilar’s CN index-based model	0.872	0.413	0.377	0.553	0.880	0.447
	Substructure-based label propagation model	0.935	0.807	0.768	0.670	0.927	0.716
	Side effect-based Label propagation model	0.936	0.809	0.771	0.674	0.927	0.719
	Off side effect-based label propagation model	0.937	0.811	0.771	0.680	0.928	0.722
	Weighted average ensemble method	0.947	0.832	0.782	0.703	0.932	0.740
	Classifier ensemble method (L1)	0.954	0.841	0.788	0.717	0.934	0.751
	Classifier ensemble method (L2)	0.952	0.839	0.784	0.712	0.933	0.746
5-CV evaluation	Vilar’s substructure-based model	0.670	0.273	0.145	0.535	0.684	0.229
	Vilar’s CN index-based model	0.872	0.413	0.377	0.553	0.880	0.447
	Substructure-based label propagation model	0.936	0.758	0.763	0.616	0.950	0.681
	Side effect-based Label propagation model	0.936	0.760	0.764	0.621	0.950	0.685
	Off side effect-based label propagation model	0.937	0.763	0.761	0.627	0.950	0.688
	Weighted average ensemble method	0.951	0.795	0.775	0.659	0.953	0.712
	Classifier ensemble method (L1)	0.957	0.807	0.785	0.670	0.955	0.723
	Classifier ensemble method (L2)	0.956	0.806	0.783	0.665	0.955	0.719

ISSN: 1471-2105