denoising with DADA2, what about the `pooling=TRUE`?

In the R DADA2's library the actual algorithm to infer the ASVs can run with three different pooling methodologies, i.e. independent, pseudo, and what I call full-pooling, i.e. the one where pool=TRUE, as shown here.

# independent
system.time(dd <- dada(drp, err=err, multithread=TRUE, pool=FALSE, verbose=0))
# pseudo
system.time(dd.pseudo <- dada(drp, err=err, multithread=TRUE, pool="pseudo", verbose=0))
# full-pooling
system.time(dd.pool <- dada(drp, err=err, multithread=TRUE, pool=TRUE, verbose=0))

On the other hand, in the manual page for denoise-paired of the Qiime2's DADA2 plugin, there is no mention concerning the full-pooling:

--p-pooling-method TEXT Choices('independent', 'pseudo')
The method used to pool samples for denoising.
"independent": Samples are denoised indpendently.
"pseudo": The pseudo-pooling method is used to
approximate pooling of samples. In short, samples are
denoised independently once, ASVs detected in at
least 2 samples are recorded, and samples are
denoised independently a second time, but this time
with prior knowledge of the recorded ASVs and thus
higher sensitivity to those ASVs.
[default: 'independent']

My question is: clearly we can't run the full pooling as we can with the R version of the package, right? Am I missing something, here?

Good morning @gabt,

You are correct: the q2-dada plugin supports many of the dada2 options and settings, but not all.

Right

That pseudo-pooling might be the best option? :stuck_out_tongue_winking_eye:

Pseudo-pooling provides a computationally efficient and scalable method to approximate the results from full pooling in linear time. In many cases, especially when samples are repeatedly drawn from the same source such as in longitudinal experiments, pseudo-pooling can provide a more accurate description of ASVs at very low frequencies (e.g. present in 1-5 reads per sample).

Of course this choice is up to you! You could perform this step in R and import your ASV tables back into Qiime2 for downstream analysis.

1 Like