Rarefying when having multiple runs

[MiriamGorostidi (Miriam)]

Thank you so much for your time!!

@llenzi

In this case the op is using PGM sequences, which may vary on length. The denoise-pyro is comparable to denoise-single, working on the full length. So it may be a good things to truncate all the sequences at the same length. The usual suggestion on not using truncation is for denoised-paired, when given the different length of ITS you could loose all samples at the merging step.
On the low quality run, I probably would try to use this to select the cut-off. Applying a shorter cut-off to the other runs should reduce the length of these sequences not the number (at least is my understanding), but of course they may loose resolution in taxonomy.

Let's say then that it is recommendable to trunc the sequences, especially truncating all of them at the same length (so I will try the analysis again using the same trunc parameter for the 3 runs, trying to maintain as much reads as I can in the lower quality sequenciation).

@llenzi

I never use PGM sequences and i am not familiar with its library preparation. What I don't know is if trimming 15 bases get rid of all the artificial sequences in the reads. Also, I wonder if using itsxpress would be helpful to normalize the sequence undergoing to taxonomy assignment at the same region in the study.

Yes, using PGM sequences is a bit rare, but it is the sequencer we have available right now... I get to the point that trimming 15 bases was enough in this discussion in the forum:

Possible Analysis Pipeline for Ion Torrent 16S Metagenomics Kit Data in QIIME2?

Hi QIIME2 community, My colleague @Lauren I have been using the Ion torrent 16S Metagenomics Kit to sequence some clinical samples that we have. Our sample sets are both longitudinal. The kit consists of two primer sets, one for V2-4-8 and one for V3-6,7-9. The products of the reactions are pooled per sample, and then the rest of library prep is completed. Each sample gets its own barcode. Ion Torrent produces single end reads. The issue here is that the reads from each of the 6 variable regio…

I suppose it would enough. However, I now have another analysis, where I know that my primers are 20nucleotides long, so, should I trim the sequences in 20?
Besides that, what is itsxpress? I have not heard about it yet.

@llenzi

When they say 'identify more taxonomies' what they meant? Less unclassified sequences or more sequences classified at species level?

I have count, for each sample, how many different taxonomies are identified (counts>0) in genera level with and without trunc. (Maybe this is not the best method...)

@llenzi

I would try blast-classifier instead of sklearn, a local alignment approach could give more coherent result across run, if they still have spurious/low quality sequences in there?

About the classifier, I have tried different ones, but I finally decided to use the feature-classifier consensus-vsearch, based on the previous Forum discussion. Should I give a try to blast?

@llenzi

On the rarefaction step, I agree with what you said, it is possible to perform it with the rep-seqs output from dada2 as most large sequences dataset and use it along all the analysis. The different ASV tables will do the selection for the relevant sequences in each case. However, they may want to compare the run after the rarefaction step, using it as kind of normalization among runs, I think.
Looking at their files, they are looking at level 6 only, they probably should look at level 4 and 5 too, to get a full idea. Form this results I would gather that trimming sequences reduce the low quality sequences, with a beneficial effect on the taxonomic resolution at genus level (at least).

This makes me more clear the rarefaction step, so thank you so much for the explanation and the discussion!
Best!!