Hi all. Firstly, thanks for a great pipeline that is so well maintained. I had a question about counting methods/modules included in the standard pipeline. Right now, the main options are RSEM and Salmon. RSEM is a standard I have been using for a while and does a great job on transcript-level abundance estimates. However, if I require simple gene-level counts, I have a couple of concerns: 1. RSEM's EM estimation of transcript abundance might be overkill in terms of complexity for gene-level counts 2. More importantly, the values are not integers. For input into something like DESeq2, integers are required. Many simply recommend rounding these RSEM values, but I feel a bit uncomfortable about that and would rather have integer gene-level counts from something like HTSeq or featureCounts, which is geared toward this problem. My question is whether either HTseq or featureCounts (or similar) were ever in the rnaseq pipeline. If so, why were they removed? Also, could you all envision putting one of these counting modules (back) into the nf-core pipeline? The modules are available, so this should be relatively easy to do, and its use could be encoded as an optional parameter or something. Lastly, I'm happy to help out integrating this if we can agree. It would be super useful for us (and hopefully others) to have gene-level counts from something fast and accurate like the two tools mentioned. Thanks!

If you wish to use HTSeq or featureCounts for quantification, you will have to either a) use an old version of the pipeline, b) create a fork of the pipeline and add the respective modules or c) manually quantify using the resulting BAM files. Once in a while this question pops up, but the reasons why it was removed are still valid. Please see previous discussions/responses in this channel, e.g. here or also @ hpatel’s Bytesize talk. PS: I am using DESeq2 all the time with the Salmon output of the pipeline. The

tximport(…, type = "salmon")

takes care of it. Also see the #C045UNCS5R9 pipeline…

Yep, the featureCounts/ HTseq way is very old-school, and the new methods do a much better job, so I don't see a reason to go back. tximport works very well for getting these results ready for downstream analysis- which is what the pipeline does when preparing its matrices. There's some good coverage in #C045UNCS5R9 about how we import the counts there.

Thanks for the quick response to both of you! Appreciate the pointers and help 🙂