Fusebench is a workbench for aggregation and interpretation of RNA-Seq gene fusions. This project was initiated as a HackSeq 2017 project, with development planned to begin during the Hackathon October 20-22, 2017.
Many cancers are defined by the presence of recurrent, subtype-defining gene fusions. While there is an abundance of informatics tools for detection of gene fusions from RNA-Seq data, these tools show little predictive overlap. Further, while annotation databases for gene fusions exist, it remains difficult to automatically annotate newly-detected gene fusions against these resources. The aim of this project is to (1) Develop methods for aggregating and comparing the results of different fusion detection tools against each other, (2) Visualizing those results in terms of fused protein domains, read evidence, and annotation status, and (3) Automatically annotating fusions for presence in selected online databases. These methods will be incorporated into a new R package and made available to the community. This project will facilitate the development of improved methods for understanding the diversity and recurrence of gene fusions, and help facilitate the clinical translation of RNA-Seq based fusion detection.
One of the goals of this project is to assist bioinformaticians and clinical scientists in interpreting the results of fusion detection tools in personalized medicine contexts. In the single-patient setting, the main goals are to filter the results of the particular analytic pipeline for high-confidence results, and then to annotate those events with event- and disease-specific information, in order to interpret the relevance of that event to a particular patients disease.
As described above, this is often difficult for RNA-Seq gene fusions, for reasons including:
- Different fusion tools often produce contradictory results from the same data sets
- Different fusion tools use different result file formats, making comparisons between tools difficult
- Useful annotation information for interpreting the likely functional effect of fusions (e.g., fused protein domains) is often not included
- Additional intra-sample features (e.g., relative expression of fusion partners) is often not included
- The process of looking up similar fusion events in online databases is often manual
- Other external data sources (e.g., fusion junction recurrence) are also difficult to link in
- Visualization of supporting evidence is also often difficult
Our ongoing planning and reference material is on the project wiki.