Assessment of computational methods for fusion transcript detection in targeted RNA-seq samples

Transcript fusions are abnormal genetic events often correlated with oncogenesis. Hence, detecting them from RNA-seq data using bioinformatics methods is an important task in cancer research. Several tools have been developed for this task, but current benchmarks are inconclusive regarding their accuracy in real data. Moreover, they have used only bulk RNA-seq samples on their assessments, and current transcript fusion detection methods have not yet been assessed under alternative settings, such as on targeted RNA-seq samples. In this article, we assess 8 transcript fusion detection tools (Arriba, ChimeRScope, CICERO, FuSeq, FusionCatcher, Pizzly, JAFFA, and STAR-Fusion) on targeted RNA-seq samples and simulated bulk RNA-seq samples. We selected the best evaluated tools in previous benchmarks. We also divided them into alignment-based and alignment-free. The targeted RNA-seq dataset comprises 92 samples of real pediatric B-cell precursor acute lymphoblastic leukemia cases with manually validated fusions. The simulated dataset comprises 10 samples with synthetic fusions. Our results point out that STAR-Fusion and FusionCatcher had superior performance on the targeted RNA-seq samples, while Arriba and STAR-Fusion had superior performance on the simulated samples. Overall, we conclude that STAR-Fusion is the most broadly robust tool across both settings.