I am very new to fusion-gene discovery and I am tasked along with some fellow colleagues to apply bioinformatic methods on clinical data for the detection of novel fusions among subtypes/groups of lymphomas patients.
We have a few sets of 2x100bp RNA-seq (~2x75Million reads) dataset and we applied FusionCatcher onto them. After some FastQC on the input reads, we found out that the first ~15 bases of each reads may be contaminated by 'not-so-random' priming or could be even be from sequencing the adapter.
We subsequently applied FusionCatcher to the hard-trimmed-first-20bps (lets call the candidate fusion from this set - B) and hard-trimmed-first-20bp-back-30bp datasets (result - C) and got different sets of results. Finally, we call the results on applying FusionCatcher with the original 2x100bp reads as A.
The absolute counts of candidate fusions in the results decrease with decreasing read-lengths. C is a power-subset of B. However, there are some detected candidate which are exclusive to B as compared with A.
After seeing this, i am wondering if there is a 'perfect' read-length to work with. (The fusions exclusive to B are under validation and i will update everyone with the results in the time to come)
What are your experiences on dealing with RNA-seq and fusion-gene finding?