Fig. 4

Performance of a machine learning classifier for fusions detected by Arriba in the BRCA and LUAD TGA cohorts. A The classifier achieved an area under the ROC curve of 0.90 and B area under the precision-recall curve of 0.83. C Predicted classification probability of each fusion event vs actual WGS validation status, at a classification threshold of 0.23. D Confusion matrix of the classifier when applied to testing data at a classification threshold of 0.23