dada2 version; learnErrors() option nreads; q2-dada2 option --p-n-reads-learn

[Phylloxera]

What version of dada2 is being used right now? I saw here that it is >1.16. However, nreads has been deprecated and replaced by nbases since version 1.7.8, but --p-n-reads-learn is still listed as an option under qiime dada2 denoise-paired --help in qiime2-amplicon-2025.7. Can I get some clarification on this apparent version/options mismatch?

[Phylloxera]

I confirmed in q2-dada2/assets/run_dada.R that nreads is indeed being applied. and it looks like it is effectively replacing the nbases option inside of dada2::learnErrors since if (!is.null(nreads) && NREADS > nreads) { break } This should mean that the qiime2 default of nreads = 1e6 is more stringent (and time consuming) than the learnErrors default of nbases = 1e8 when the average read length is > 100 and less stringent when the average read length is < 100. Maybe @benjjneb switched to nbases because it removes the effect of read length on the error calculation?

[benjjneb]

Maybe @benjjneb switched to nbases because it removes the effect of read length on the error calculation?

Correct.

[lizgehret]

Hey @benjjneb, if our understanding is correct it looks like nreads is still a usable param (we call it here in q2-dada2) even though it's been deprecated. We are using bioconductor-dada2=1.30.0 which is well past the version where nreads was deprecated. Would you recommend we update our usage of nreads to nbases? If so, when do you expect to fully remove the functionality for nreads?

[benjjneb]

If so, when do you expect to fully remove the functionality for nreads?

I don't intend to remove it.

Would you recommend we update our usage of nreads to nbases?

I think it's a low priority. The main use case is for long-read sequencing, which is already broken out into a separate q2 action. Leaving nreads as is for the denoise-single and denoise-paired commands intended for short-read data is fine.

[Phylloxera]

If an nbases option is added and nreads are allowed to take the value NULL, then results can be repeatable between dada2 (when using the default with nbases) and q2-dada2. Otherwise, not.