Fig. 3

Cross validation of gene-function prediction. a Average ROC for edge prediction (TPR: true positive rate, FPR: false positive rate). b Average precision-recall curve for edge prediction indicating absolute z-score cut offs for different precision levels. c AUC for edge prediction as a function of the embedding dimenion K. The optimal embedding dimension is \(K\approx 100\), smaller values of K lead to underfitting, while greater values lead to overfitting. d Average ROC for edge sign prediction. e Average precision-recall curves for edge sign prediction (blue: prediction of activation vs. inhibition, orange: prediction of inhibition vs. activation). Error bars and grey areas around curves represent standard deviations estimated from replicated tests