IPMiner: hidden ncRNA-protein interaction sequential pattern mining with stacked autoencoder for accurate computational prediction

Table 3 Performance comparison on non-structure-based NPInter2.0 and RPI13254

Dataset	Method	Accuracy	Sensitivity	Specificity	Precision	MCC	AUC
NPInter2.0	IPMiner	0.952	0.946	0.959	0.945	0.904	0.995
	SDA-RF	0.937	0.940	0.935	0.941	0.876	0.975
	SDA-FT-RF	0.934	0.953	0.912	0.955	0.868	0.990
	RPISeq-RF	0.944	0.940	0.949	0.940	0.889	0.978
	lncPro	0.928	0.919	0.938	0.917	0.856	0.971
RPI13254	IPMiner	0.945	0.905	0.995	0.895	0.896	0.985
	SDA-RF	0.699	0.717	0.658	0.741	0.400	0.761
	SDA-FT-RF	0.813	0.728	0.998	0.626	0.675	0.901
	RPISeq-RF	0.739	0.766	0.688	0.790	0.480	0.817
	lncPro	0.712	0.716	0.701	0.723	0.424	0.792

For RPI13254, it has 13524 positive pairs and 5172 negative pairs. Here we randomly sub-sampling positive pairs from original paper to create balanced dataset, so it actually consists of 5172 negative pairs and 5172 positive pairs
The boldface indicates this measure performance is the best among the compared methods for individual dataset

ISSN: 1471-2164