Table 7. Performance of meta-predictors using preprocess-I transformation under multi-fold cross validation and in independent dataset.

Predictor	Dataset	SENS	SPEC	ACC	MCC
Meta-I-4S	D163*	93.3 ± 2.8%	98.8 ± 1.6%	96.1 ± 2.2%	0.92 ± 0.04
	D1679**	88.1 ± 2.4%	92.9 ± 0.1%	89.5 ± 1.7%	0.76 ± 0.03
Meta-I-5L	D1679*	79.4 ± 4.4%	94.0 ± 1.3%	86.6 ± 1.6%	0.74 ± 0.03
	D163**	98.9 ± 0.4%	93.2 ± 3.3%	96.0 ± 1.4%	0.92 ± 0.03

(*) Meta-I-4S is composed of four individual predictors: MiPred, miReNA, MiRPara, and ProMiR. The predictor was optimized in the D163 dataset using three-fold cross validation; Meta-I-5L is composed of five individual predictors: MiPred, miReNA, MiRPara, ProMiR, and TripSVM. It was trained in the D1679 dataset using five-fold cross validation.

(**) The performance of these two predictors in independent dataset, which was D1679 for Meta-I-4S and D163 for Meta-I-5L, was averaged over three- or five-iterations of prediction that correspond to three- or five-fold cross validation. Errors were standard errors calculated from either three- or five-iterations of prediction.