Fig. 5

Decision set models performance. Two DS models have been evaluated: one trained with all valid paths (DS incl.Pheno) and one trained without Phenotype-traversing paths (DS excl. Pheno), on a stratified 10-fold cross-validation setting. A test set of 15 pathogenic gene pairs from recent literature has been held out for independent evaluation. A Receiver operating characteristic (ROC) curve obtained by averaging all fold curves. The best classification threshold is evaluated using the geometric mean between the sensitivity and the specificity. B Precision-Recall curve obtained by averaging all fold curves. C Test set predicted probabilities, displayed in decreasing order. The horizontal lines represent the optimal thresholds for binary classification