Fig. 4

Comparison of CICERO with other methods on driver fusion detection. a Distribution of leukemia, solid tumor, and brain tumor in the 170 RNA-seq used for benchmark test. b Prevalence of recurrent (≥ 3) gene fusions in the benchmark data sets stratified by the following four classes: chimeric transcript caused by exon-to-exon fusion expressed at high (> 5 FPKM) or low level, internal tandem duplication (ITD), and other non-canonical fusions involving intronic or intergenic regions. c Comparison of the sensitivity (top panel) and ranking of the driver fusions among all predicted fusions (bottom panel) by CICERO and five other methods (ChimeraScan, deFuse, FusionCatcher, STAR-Fusion, and Arriba) in the four categories of driver fusion. The ranking by CICERO, labeled CICERO_raw, is based on fusion score alone without incorporating matches to known fusion status. Error bars representing standard deviation of detection sensitivity at the top panel were calculated by bootstrapping of samples with 100 iterations. d True positives (dark blue) and false positives (light blue) of predicted somatic fusions identified by CICERO and other fusion detection programs. The exact number of events is marked as (true positive/total prediction) under the name of each method. CICERO’s high-quality predictions are compared to those of STAR-fusion and Arriba (left panel) while all CICERO predictions are compared to those of FusionCatcher, deFuse, and ChimeraScan (right panel)