Fig. 4

Application of structure scores for structure prediction using a machine learning approach and in silico pseudo-energy model [10] for the test set of three rRNAs. a Stem/loop structure prediction accuracy based on the structure scores computed from individual or combined datasets (icSHAPE in vivo and in vitro and PARS datasets) by the support vector machine, with the removal of indeterminable scores. Average and variance of accuracy were computed after 10-fold cross-validation. The predictions based on a single dataset utilized the structure scores from either in vivo icSHAPE dataset, in vitro icSHAPE dataset, or PARS V1 dataset, and the results of the maximum mean accuracy are presented. b Positive predictive value (PPV) and sensitivity of in silico structure prediction using RNAfold with and without reactivity assisting. The best PPV and sensitivity pair was selected to maximize their sum for each prediction condition