Strand cross-correlation analysis before, or after de-duplication? about chipseq

[apeltzer]

This looks pretty related to that: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3989762/

[apeltzer]

"Duplicated reads can be seen to contribute to both artifact and ChIP signal and the removal of duplicated reads significantly reduces ChIP signal across samples. The evaluation of ChIP quality following duplicate removal may therefore underestimate the extent of ChIP enrichment relative to background and so careful consideration of the contribution of duplicates to artifact regions and ChIP signal must be made prior to evaluation of NSC and RSC metrics."

[ewels]

Yup, that's the same publication that's linked in the original comment I think.

I'm not keen on using dedup data for some steps and not others, as originally suggested by @sifakise. But we could certainly have a --nodedup flag to skip the deduplication.

We're calculating NSC and RSC in the pipeline, as well as collecting % duplication stats, so then it would be up to the user to choose whether to do it or not.

Phil

[apeltzer]

Oh, sorry I didn't spot that :-(

I'd also say keeping things consistent is better and if a user knows whether dedup makes sense they should be able to set that with a parameter as you suggested.

[chuan-wang]

Fixed in PR #38