I opened a PR to add decontam to the pipeline: https://github.com/nf-core/ampliseq/pull/948 Feedback welcome!

I had a look and left only a comment for now. Let me know when you want a full review.

Thanks! Questions: • About better documentation: I could add a paragraph just before https://nf-co.re/ampliseq/2.16.1/docs/usage/#taxonomic-classification about decontamination/decontam, or is there a better place? • What do you think about the sequence of events? Right now decontam after ASV generation with DADA2: VSEARCH -> decontam -> barrnap -> length filter -> codon filter -> ITSx -> taxonomic classification (all optional...). Is decontam at the right place? I thought its best do add it as early as possible, but after clustering with vsearch, do you agree? I have only minimal experience with decontam unfortunately.

What I was thinking specifically about in documentation is the

quant_reading

field in the input sheet.

Yes, I see, I need to put that indeed in a proper documentation.

I'm confused by Vsearch -- is that an input to Decontam?

So after the ASVs are produced by DADA2, the pipelines offers various tools to reduce ASVs before taxonomic assignment. The first tool is VSEARCH for clustering ASVs to OTUs. In my PR, the next tool is decontam, so yes, the input is from VSEARCH or (DADA2 if OTU clustering is skipped). But generally, as a tool, decontam can take any abundance table and is not related to VSEARCH in any way.

I could imagine that Decontam would be stronger if artifacts were already removed so the program only sees true target sequences. Moreover, if a user were to apply Decontam themselves, they would likely do so after any filtering by the pipeline.

So I could move Decontam as last in the sequence, instead of first. You assume that this increases sensitivity, or do you know that? I tend to agree, but I dont know.

Also, assigning taxonomy to all ASVs, not only those that pass Decontam would make sense to me.

Currently in the PR, whenever the required column(s) are provided in the samplesheet, Decontam is run but nothing is filtered. That means, all ASVs are going into taxonomic classification. Decontam annotation is just treated as an information, not more. However, when -

-decontam decontaminate

or

--decontam notcontaminant

is used, then the correspoding Decontam results are applied to remove ASVs. My intention here was to be able to skip contamination and reduce runtime with taxonomic assignment, if desired. And to subsequently use QIIME2 visualisations on filtered ASVs. .

OK. Then the only thing that matters might be what the user expects. And I can't say I have a strong feeling for what that might be. Easiest to keep the way you've already implemented it perhaps?

Sure, its easiest to keep it as is 😄 But otherwise it would be just copy paste to another place since the code is quite clean in that section of the script I think. I'm going to improve documentation and then mark the PR as ready.

I agree: No feature without a test! I guess Decontam is using random numbers -- did you set the seed?

Uff, good point, no seed yet.