Star-fusion not finding reference fusion genes

Question

[psaundphd]

I am trying to qualify Star-fusion on Galaxy for doing cancer gene fusion studies. I downloaded a complete set of CTAT files for running it and used beautifully clean validation RNAseq SRR samples from PRJNA436593 designed for the purpose. It failed to find expected NPM1-ALK fusions in four different cancer types which found it by Tophat-fusion. It also failed with finding the BCR-ABL1 fusion in CML, which is the diagnostic marker for the cancer. I used the default settings in the Galaxy installed instance of Star-fusion.

Any suggestions on how to change the settings in order to make it find the reference fusions? If it doesn't find the expected ones how can I have confidence in its ability to find unknowns?

Answer 1

[Kevin Blighe]

Check, for example, the cut-off for read depth at the breakpoint. Also check for strand information. Essentially, go through each parameter that's available to you, and compare these to the equivalent parameters in Tophat-fusion.

The detection of fusion genes from NGS data, like copy number variants/alterations, suffers from a lack of reproducibility - NGS data is 'messy' and poor quality, generally speaking. This lack of reproducibility then, I believe, results in a large number of programs being developed, each claiming to be better than the next and invariably tested on different datasets. The same is true for predicting functionality / pathogenicity of variants.

For fusion gene detection, take a look: A: Gene Fusion Detection: Rna-Seq Data

Kevin