Two-pass on the fly vs regenerated genome method with SJ filtering

[FrancescaBC]

Hi everyone,

I've just started working on including alternative splicing into our current rnaseq pipeline and I'm seeing different opinions out there on how best to incorporate the STAR two-pass mode.

In the STAR manual, for a per-sample basis, using the twopassMode Basic is recommended over the older method with the re-generated genome.

However, I have come across some online pipelines which use the older method, but then include some filtering steps on the SJ.out.tab file (e.g. filter out SJ which is a noncanonical motif and is only supported by < 3 unique mappings) to get a more high confidence list of SJ which is then incorporated into the second pass.

Does anyone have some insight on how the older method with filtering would compare with the newer on the fly 2-pass method? In my mind, using a filtered list of higher confidence SJ sites would improve the overall performance and ultimately improve the downstream alternative splicing detection methods (e.g. RSEM-turbo).

Thanks in advance for sharing your experiences and insights on this. I'm particularly curious if anyone has compared their alternative splicing outputs using BAM files generated between these two methods, or if there are specific scenarios where one might outperform the other. Any advice or examples from your own pipelines would be greatly appreciated.

Looking forward to learning from this community!

Best regards,

Francesca BC