Interpretation of results after taxonomy analysis from decontaminated table

[joaomiranda (João Miranda)]

Hi @Mehrbod_Estaki ,

Thanks a lot for your observations! I'll put it into practice and return later with better results.
But before that, I would like to clear some doubts:

@Mehrbod_Estaki

I don't see cutadapt being used in the provenance of your table-single2.qzv, are you sure you used the right input file there?

I'm sorry this is the wrong input, I've runned dada2 without the output from cutadapt, I'll run correctly now.

I've runned cutadapt with --p-discard-untrimmed:

qiime cutadapt trim-single \
 --i-demultiplexed-sequences single-end-demux.qza \
 --p-front CCTACGGGNGGCWGCAG \
 --o-trimmed-sequences trimmed_seq_single.qza \
 --p-error-rate 0 \ 
 --p-discard-untrimmed \
 --verbose

qiime demux summarize \
  --i-data trimmed_seq_single.qza \
  --o-visualization trimmed_seq_single.qzv

Here you can visualize my quality plot, and I have some questions about the trim and trunc parameters after cutadapt: trimmed_seq_single.qzv (296.1 KB)

@Mehrbod_Estaki

Next, in your first run you trimmed 65 nt from your 5' and truncate at 240 from your 3'. In the second run you actually trim less from the 5' with 40 and truncate to the same 240 spot. From looking at your quality plots I'm not surprised you're seeing such a massive drop at the initial filtering step of DADA2. Unfortunately your quality scores are quite low on your 5'. If I were you I would trim at least 51 nt to get you passed that initial dip in quality scores, I'd also try a second run with 80+ nt trimming to see you pass that second odd dip too. Luckily for you even if you were to trim that much from the 5' you should still capture a bunch of the V4 region and that should still give you decent resolution.

Based on the middle of the box in the positions of the quality plot at my 5', we can see a low Qscore in 43 and 44 nt but in the 79 nt, where are other odd dip, we can see a Qscore 38 looking for the middle of the box. So based in this observations and looking for the quality in the 3' I thought about running --p-trim-left 45 and --p-trunc-len 235 , is that correct?

@Mehrbod_Estaki

There's a previous discussion on this if you want to see an example of what to do here. You can use much lower % identity and alignment than what I did a few years ago in that post. The idea is to just filter your reads to toss away anything that doesn't look like 16S, we're not looking for perfect matches at that step.

Thank you so much! I'll take a look in this discussion and try to apply in my data

@Mehrbod_Estaki

The second I noticed was that when you are extracting your reads for training the feature-classifier, you are truncating your reads at 200 with no trimming, however your dada2 parameters have you doing a 40/240 trim/truncate. What is essentially happening here is that the region you have extracted and trained your classifier is different than the ones where your ASVs represent. So, as @Keegan-Evans already mentioned:

I don't understand my error here, I based my truncating in the extract-reads based on this discussion. I used the truncating parameter based on the length of my resulting amplicon (240-40 = 200) and set --p-min-length and --p-max-length to 0. How can I identify the region where my ASVs are correctly represented? I have to insert --p-trunc-len and --p-trim-left in my code with the same parameters as I used in dada2?

Thanks in advance!