Dada2 error with grinder simulated sequence errors

[Nicholas_Bokulich]

Hi @Jeremie_Auger,

@Jeremie_Auger

incorrect number of subscripts on matrix

This occurs when dada2 encounters artificial quality scores, as you have discovered, and as described here:

Doing taxonomy analysis and getting abundancies with manifests

If I’m reading the provided sample correctly, this appears to be single-end 454 data with fake quality scores (everything is Q=40). Is that correct? The DADA2 plugin will fail on such data, as it is expecting a range of scores as in real data. While you can run DADA2 on 454 data, it needs the real quality scores. (Note to self: add flag to plugin to allow running w/o quality scores). I am not a Deblur expert, but if this is 454 data you might want to double-check with @wasade on the right way …

@Jeremie_Auger

After adding some “sequencing errors” to my in silico fastq, I have been able to use Dada2, but only with the parameters that are in the moving pictures tutorial

How many errors are you inserting and how? What does the quality score distribution look like on your sequences? I strongly suspect dada2 is still failing because you are giving it artificial-looking quality scores, they are still breaking the assumptions of this method.

I think the bottom line is that if you want to use dada2, you need to give it the intended form of data: sequences with at least realistic quality scores.

@Jeremie_Auger

at least, Dada2 doesn’t discart representative sequences at is seemingly happens with Deblur).

That is the purpose; deblur identifies sequences that look noisy and throws them away. dada2 does the same if the reads look too noisy, and the fact that you get no sequences filtered out with dada2 is probably a good indication that your quality scores are not "real" enough. The fact that you do not want reads discarded leads me back to my earlier question: do you really want to denoise these data? maybe you do not want to "noise" it to begin with.

EDIT: I see now that you declared the purpose earlier on:

@Jeremie_Auger

I have issue running Dada2 on an artificial in silico dataset of a 10 species mock community. (My aim being to compare those perfect fastq to the actual mixes we made and sequenced v3-v4 on a MiSeq).

So I don't think you want to generate noisy data to begin with, if you want to compare the sequenced compositions to the expected compositions. You can just compare the real to the "perfect" composition downstream as a feature table. I have a little example of how to do it in this tutorial (ignore the quirks related to fungi, skip to the part about q2-quality-control and the various methods in there that you can use):

Fungal ITS analysis tutorial

Fungal ITS analysis, mock communities, and more fun NOTE: This tutorial was written in a QIIME 2 2018.11 environment. It is not guaranteed to work with earlier or later versions of QIIME 2. This tutorial was compiled as a working exercise for a QIIME 2 workshop in December 2018, and does not represent the only possible fungal ITS workflow with QIIME 2, or even a benchmarked protocol recommendation. See other tutorials, e.g., the q2-itsxpress tutorial for other fungal ITS analysis options in QI…

I hope that helps!