Fig. 4

Fusion prediction accuracy on 56 cancer cell lines. a The distribution of leaderboard rankings for accuracies assessed using the varied truth sets. Methods are ranked from left to right according to median accuracies. b The distributions of execution times for all cancer cell lines are shown. All methods were run on the Broad Institute computing grid with commodity hardware and allocated single cores, with the exception of the two slowest methods, TrinityFusion-UC and TrinityFusion-D, which were each given four cores. c Median rankings are plotted vs. median run times, with a black dashed box drawn around the most accurate and fastest methods. d The PPV and TPR are shown at maximum point accuracy (F1) for an example trial involving the truth set defined as requiring at least seven methods to agree. The most accurate methods are found to cluster into groups of high sensitivity (top dashed rectangle) or high precision (right dashed rectangle)