LBSizeCleav: improved support vector machine (SVM)-based prediction of Dicer cleavage sites using loop/bulge length

Table 3 Results on average specificity, sensitivity, accuracy, and MCC for both 5p and 3p arms by five-fold cross-validation using PHDCleav and LBSizeCleav (k=1,⋯,5) with window sizes 8,10,12,14 based on secondary structures predicted by RNAFold

Method	Window size	5p arm				3p arm
		Sn	Sp	Ac	MCC	Sn	Sp	Ac	MCC
Extended PHDCleav	8	0.746	0.744	0.745	0.490	0.772	0.750	0.761	0.522
	10	0.792	0.783	0.787	0.575	0.779	0.800	0.790	0.580
	12	0.798	0.799	0.798	0.597	0.785	0.830	0.808	0.616
	14	0.778	0.813	0.795	0.591	0.805	0.853	0.829	0.659
LBSizeCleav (k=1)	8	0.739	0.805	0.772	0.545	0.785	0.790	0.788	0.576
	10	0.798	0.820	0.809	0.618	0.795	0.815	0.805	0.610
	12	0.792	0.815	0.803	0.607	0.822	0.840	0.831	0.662
	14	0.815	0.822	0.819	0.638	0.851	0.852	0.851	0.703
LBSizeCleav (k=2)	8	0.753	0.788	0.771	0.542	0.792	0.788	0.790	0.580
	10	0.816	0.795	0.806	0.612	0.811	0.794	0.803	0.606
	12	0.836	0.784	0.810	0.621	0.814	0.803	0.808	0.617
	14	0.845	0.769	0.807	0.616	0.867	0.800	0.834	0.669
LBSizeCleav (k=3)	8	0.751	0.794	0.773	0.546	0.784	0.797	0.790	0.581
	10	0.808	0.808	0.808	0.615	0.795	0.813	0.804	0.608
	12	0.822	0.800	0.811	0.623	0.808	0.835	0.821	0.643
	14	0.816	0.803	0.809	0.619	0.853	0.838	0.846	0.692
LBSizeCleav (k=4)	8	0.764	0.772	0.768	0.536	0.809	0.772	0.790	0.581
	10	0.824	0.762	0.793	0.587	0.824	0.766	0.795	0.590
	12	0.841	0.737	0.789	0.581	0.842	0.756	0.799	0.600
	14	0.871	0.678	0.774	0.559	0.898	0.697	0.797	0.607
LBSizeCleav (k=5)	8	0.782	0.747	0.764	0.529	0.822	0.744	0.783	0.568
	10	0.836	0.732	0.784	0.572	0.829	0.726	0.777	0.558
	12	0.867	0.699	0.783	0.574	0.864	0.682	0.773	0.556
	14	0.899	0.626	0.763	0.546	0.917	0.619	0.768	0.562

Sn, Sp, Ac, and MCC denote sensitivity, specificity, accuracy, and Matthews correlation coefficient, respectively

ISSN: 1471-2105