Fig. 7

The relationship between PRDIS specific predictors and reference methods for proteins in VS2168 dataset. In this figure we compare the performance of PRDIS specific methods when applied to the variants in VS2168. None of the proteins represented in this set contributes a variant to VS228, which is the dataset used to train the specific predictors and obtain a cross-validated estimate of their performance (Fig. 4). The MCCs of the specific methods are represented with violin plots and those of the reference methods (SIFT, PolyPhen-2, PON-P2, CADD and MutationTaster2) are represented with black boxplots. We see that, in spite an overall decrease in performance for all tools displayed, specific methods are frequently better than reference methods, but there is an increasing overlap between both approaches as the performance of the reference method grows (e.g. in the cases of PON-P2 or PolyPhen-2). This result confirms the conclusion obtained with the VS228 set (Fig. 4)