👋 Hi everyone, following the discussion in PR #260, we’re considering a possible change in how contrast files are handled in the

nf-core/crisprseq

pipeline, for the screening analysis. Currently, the contrast file can refer either to conditions or to samples, depending on the context (e.g. when using precomputed count tables). One proposed simplification would be to remove the “condition” column entirely and make the contrast file always refer to samples. This would make the behavior more consistent across cases, though it might slightly reduce readability and would be a breaking change compared to previous releases. What do you think? Please vote with an emoji or share your thoughts below! • 👍 = I support simplifying and using samples only • 🤔 = I’m unsure / need to understand implications • 👎 = I’d prefer keeping the current approach

Hi 👋 I'm the one who opened the original issue on this. In my opinion as a user of multiple nf-core pipelines, I would mirror the behavior of samplesheets in #CE8SSJV3N and others as much as possible. To me that means maintaining both a

sample

and

condition

column, that way individual replicate samples can be properly labeled in the output and the cross condition comparison properly interprets the contrasts. I understand and appreciate the desire to simplify, but I'd vote to keep the samplesheet the way it is and change how they are handled under the hood

Wouldn#t that complicate running the pipeline from count tables rather than raw fastq files?

@Joachim Fritscher with the proposed change, we would keep the input from count table as it is (when running from count tables, you currently do not have conditions but just sample labels, and you define contrasts by sample labels). On the contrary, if we keep the conditions, we feel that we need to make changes to the way that the input is given when running from count tables, making sure that in all the cases the contrast file contains condition labels.

Hi, sorry didnt't write my vote reason. My hesitation is for chemogenetic CRISPR screen, especially drugz. I am the one who add the drugz module to the pipeline and sometime people may need to use different contrasts different time of the drug introduced which this help them sort the contrast better. I didnt investigate this throughly in the script wise but this could affect that badly (this include their mageck rra and bagel2 analysis too)

Thanks for the clarification. Just as a side note, this concern should now be addressed: since PR #252, the contrasts are internally translated to sample IDs. This ensures that the output matrices from

mageck count

consistently report sample IDs. Either choice, using contrasts defined either by conditions or by sample ids, can now be easily implemented in the downstream analyses (including MAGeCK RRA, BAGEL2, or drugz).