A single network comprising the majority of genes accurately predicts the phenotypic effects of gene perturbation in C. elegans

SUPPLEMENTARY FIGURE S1 ROC plots illustrating the Wormnet-based prediction of RNAi phenotypes. For each known phenotype, we analyzed the ability to predict genes conferring the phenotype using leave-one-out analysis. Every gene in the Wormnet was first rank-ordered by the sum of its LLS scores to all other genes with the given RNAi phenotype; we then measured the recovery of genes with the given phenotype, calculating true positive rate (TP/(TP+FN)) and false positive rate (FP/(FP+TN)) as a function of rank. In each plot, the diagonal represents no predictive power, curves above the diagonal indicate prediction of the plotted phenotype, with curves farther to the top left of the plot indicating the strongest predictive power. In order to measure rates up to 100%, we employed pseudocounts, assigning a very low LLS score (0.00000001) to all unlinked gene pairs in Wormnet (i.e., gene pairs lacking all evidence for functional coupling).