Table 3:

Performance of different machine learning classifiers on the training dataset S2′ via 5-fold cross-validation

Classifier	Method	Features	Threshold dependent					Threshold independent
			Score range	Thresholda	Sensitivity	Specificity	MCC	SN_496b	AUC
Basic	KNNc	518	0.100∼0.900	0.500∼0.550	0.593	0.621	0.214	0.607 ± 0.014	0.6305
	RFd	Random	0.080∼0.900	0.380∼0.579	0.590 ± 0.168	0.617 ± 0.183	0.219 ± 0.019	0.600 ± 0.007	0.6413 ± 0.0082
	SVM	518	0.328∼0.743	0.542	0.567	0.681	0.250	0.615	0.6509
Optimised	SVM + FFS	78e	0.170∼0.836	0.561	0.518	0.760	0.287	0.621	0.6768
	SVM + ASI	74e	0.098∼0.918	0.549	0.623	0.750	0.376	0.681	0.7479

^aThis threshold is provided by maximising the value of MCC.

^bThis sensitivity is measured among tested genes with the top 496 prediction probabilities.

^cThe k-value here is set as the square root of the size of the training samples in 5-fold cross validation (i.e., k = 20) [62].

^dThis random forest algorithm uses 50 random grown trees and the modelling and validation procedures are repeated 10 times.

^eThese features constitute the best/optimum feature set for the current machine learning method.