Figure 4

Comparison to existing methods at site level. ROC curves for site-level mirMark and existing methods on the hold-out test set using different overlap thresholds to the expected target regions: (A) 25%, (B) 50%, (C) 75%, and (D) 95%. (E) The AUC of the ROC curves from false positive rate 0 to 0.4 (chosen based on results from a-d) under different overlap thresholds. (F, G) The number of predicted regions overlap the expected regions at about 0.6 true positive rate for (F) mirMark random forest, (G) SVMicro, and (H) MiRanda.