Comparative assessment of methods for the fusion transcripts detection from RNA-Seq data

Abstract

RNA-Seq made possible the global identification of fusion transcripts, i.e. “chimeric RNAs”. Even though various software packages have been developed to serve this purpose, they behave differently in different datasets provided by different developers. It is important for both users, and developers to have an unbiased assessment of the performance of existing fusion detection tools. Toward this goal, we compared the performance of 12 well-known fusion detection software packages. We evaluated the sensitivity, false discovery rate, computing time, and memory usage of these tools in four different datasets (positive, negative, mixed, and test). We conclude that some tools are better than others in terms of sensitivity, positive prediction value, time consumption and memory usage. We also observed small overlaps of the fusions detected by different tools in the real dataset (test dataset). This could be due to false discoveries by various tools, but could also be due to the reason that none of the tools are inclusive. We have found that the performance of the tools depends on the quality, read length, and number of reads of the RNA-Seq data. We recommend that users choose the proper tools for their purpose based on the properties of their RNA-Seq data.

Using the Seven Bridges Cancer Genomics Cloud to Access and Analyze Petabytes of Cancer Data

“…The CGC is well suited to addressing such a problem, because multiple algorithms and datasets are easily available and thus each gene fusion caller can be rapidly and directly benchmarked against others. For this example, we will demonstrate how to compare the fusion callers EricScript, FusionCatcher, and STAR-Fusion on the CGC, based on the example in (Kumar, Vo, Qin, & Li, 2016). …”

Recurrent and pathological gene fusions in breast cancer: current advances in genomic discovery and clinical implications

“…In addition to being relatively less expensive with a quick turnaround time, it can also quantify expression levels and facilitate the detection of multiple fusion variants generated during the fusion event, making this a more ideal technology for gene fusion detection. However, RNA -seq routinely generates a daunting quantity of chimeric sequences, most of which are artifactual fusion sequences as a result of library artifacts and mapping errors [21,22], transcription-induced chimeras (TIC) resulting from intergenic splicing of adjacent genes, or passenger gene fusions. Bioinformatics methods that use stringent parameters to filter out artifactual chimeras may become less sensitive in detecting authentic fusions or underestimate their incidence, while filtering out the chimeras from adjacent genes will dismiss the close-range gene translocations.…”

Untitled

“…However, there is still opportunity for performance improvement. Kumar et al [14] have shown that when comparing 12 different software packages they observed performance dependency in quality, read lengths, and supporting reads. More recently, Haas et al [15] have compared 16 different packages to assess fusion prediction.…”