Comparison of PPVs for various classifiers. Positive predictive value for SVM classifiers trained on combined data with and without using 2Hop characteristics as inputs. At a threshold level of 0.999, the classifier trained with all graph theoretic protein properties and 2Hop SS as inputs has the highest positive predictive value. Addition of the extra indicator variable 2Hop SP as input results in a classifier with decreased positive predictive accuracy. The subplot in the figure shows the number of predicted pairs (true positives and false positives) at different levels of threshold.