Remove secondary alignments in RNA-seq analysis?

Question

[nanoide]

Hello, so I'm mapping reads from a RNA-seq library to a reference genome using hisat2 with default parameters. Of my total output mapped reads, when checking with samtools flagstat/stats I'm getting a significant amount of secondary alignments (SAM flag 0x100) I've looking on how to deal with these. I have read a good practice article and it says I should keep them, but others keep only uniquely mapped reads and recommend to remove them. Any thoughts? Tools in downstream analysis, for example htseq-count are going to be able to manage them correctly right? I would lose too many reads by removing them, so it is correct if I kept them?

Thank you for time!

[nanoide]

Thanks for your answer, I'll keep looking just to be sure I'm proceeding correctly, but what you mentioned is the case.

Answer 1

[Devon Ryan]

Since it sounds like you'll be extracting counts with featureCounts or htseq-count then you can leave the secondary alignments in, as both of the aforementioned programs will ignore secondary alignments (and their associated primary alignments).

[rsafavi]

What if you want to assemble the transcript? Do you still suggest to keep or remove the secondary alignments and the supplementary?

[Devon Ryan]

You'll need to keep at least the primary alignments, since that'll break things like stringTie otherwise.