Meta analysis with different 16S region and get too low features after DADA2

[jwdebelius (Justine Debelius)]

Hi @Jonathan,

I am so sorry for disappearing on this.

You pipeline mostly makes sense, but I think there are a few checks you may want to revisit

@Jonathan

1. Import fastq (individual study, respectively)
2. cut primer sequences with qiime plugin referring to their papers

Are you discarding untrimmed reads when you do primer trimming? Have you checked how many reads you're losing at this step? I would recommend making sure you're not discarding everything if the primers have already been trimmed.

@Jonathan

3. as one of studies are single ended, I decided to use only forward read for other studies

I would recommend this for a primer pair, but not for all the studies overall.

@Jonathan

4. quality control with qiime plugin
5. denoise with dada2 single with same trunc length parameter (qiime dada2 denoise-single) without --p-max-ee parameter

You should not do quality control before you run DADA2. DADA2 assumes that your reads not quality filtered, so I would recommend either using Deblur if you want to quality filter, or not doing that step. I would recommend the same parameters within the same hypervariable region, but would suggest that you can do variable regions specific primers.

I think the fact that you're tracing back to the ASV table is smart, and would recommend checking your read statistics in each step.

@Jonathan

6. OTU clustering (closed reference) with silva-SSU-99 reference with --p-perc-identity 0.97 parameter

I again think this is smart it lets you scaffold all the sequences using consistent identifiers.

@Jonathan

After taxonomy analysis with OTU data, taxonomy is not assigned on the 2 studies while other studies are well assigned with silva-full-length-pre-trained classifier

I don't understand this step. Why are you doing taxonomic assignment after you did closed refernece OTU clustering? One of the beauties of closed reference clustering is that the taxonomy is already assigned and the tree is already built for you. You just need to import them into QIIME 2 and you're ready to go!

So, I think I'd suggest the following modifications

• Check your sequences after primer trimming for each study to make sure that reads have not been discarded
• Use paired end reads for regions where you have paired end reads (except V13, that needs to be single end if its Illumina)
• Either perform quality filtering and denoise with deblur or use DADA2 alone
• Don't try to do taxonomic classification
• Keep checking the number of reads retained at each step
• Keep using consistent parameters within a single region
• Keep asking questions.

Best,
Justine