. 2012 Feb 16;7(2):e31791. doi: 10.1371/journal.pone.0031791

Table 2. CP Viability Prediction Performance of Various Procedures.

Dataset	Performance measure	Closeness+^a	Farness: F_b+^a	HI	ANN	RF	SVM	Combined^b
Dataset T^c	AUC	0.746	0.768	0.828	0.885	0.844	0.819	0.905
	Sensitivity	0.577	0.761	0.771	0.852	0.775	0.647	0.857
	Specificity	0.677	0.569	0.723	0.846	0.714	0.762	0.790
	False positive rate	0.323	0.431	0.277	0.154	0.286	0.238	0.210
	MCC	0.264	0.329	0.490	0.690	0.483	0.407	0.632
DHFR	AUC	0.814	0.873	0.843	0.833	0.840	0.822	0.906
	Sensitivity	0.465	0.849	0.616	0.593	0.733	0.605	0.709
	Specificity	0.918	0.740	0.918	0.863	0.808	0.918	0.918
	False positive rate	0.082	0.260	0.082	0.137	0.192	0.082	0.069
	MCC	0.421	0.594	0.551	0.467	0.539	0.541	0.633
nrCPDB-40	Sensitivity	0.622	0.616	0.733	0.735	0.733	0.778	0.746
nrGIS-40	Sensitivity	0.614	0.590	0.700	0.682	0.698	0.715	0.715

Random forest was applied in this experiment to the assess the prediction power of closeness and farness.

A combination of the four machine learning methods (HI, ANN, RF and SVM) by averaging their probability scores into a single score. See the main text for details.

These results were obtained with 10-fold cross-validation.