Additional file 4 of Finding functional associations between prokaryotic virus orthologous groups: a proof of concept

Additional file 4: Fig. S3. ROC curves for the 10 datasets (N1–N10) used for performance evaluation. Each dataset was used as a ground truth set for parameter optimization of a Random Forest classifier (70% training). The resulting best model was used for predictions on the holdout set (30% of the original) and its ROC curve is depicted by a thicker line. The remaining 9 datasets were used as input for the best model obtained for training (70%) and holdout (30%) and their ROC curves are shown as more transparent lines (AUROC = Area Under the Receiver Operating Characteristic)