I am analyzing Oxford Nanopore 16S data from gastric biopsies of patients positive for Helicobacter pylori by histology.
I initially had 24 samples from 12 patients: one antrum biopsy and one corpus biopsy per patient.
The raw sequencing output contained 28,752 reads. After filtering with NanoFilt using Q7 and minimum length 200 bp, I imported 23,784 reads into QIIME 2.
My pipeline was:
Dereplication with VSEARCH
De novo OTU clustering at 97%
Chimera removal with UCHIME de novo
Filtering rare features with min-frequency 2
Taxonomic assignment with VSEARCH against SILVA 138.2
Removal of contaminants/unassigned features
Diversity analyses after rarefaction
Final table:
23 samples
1,231 OTUs/features
6,258 observations
rarefaction depth: 130 reads/sample
Taxonomy with VSEARCH/SILVA 138.2 detected:
10 phyla
91 genera
48 species-level taxa
My concern is that 1,231 OTUs/features and 91 genera seem high for gastric biopsy samples with relatively low read depth.
My questions are:
Does this final number of OTUs/features seem plausible for Nanopore 16S data, or is my pipeline still too permissive?
Is 97% OTU clustering with min-frequency 2 reasonable here, or should I use stronger filtering?
Would you recommend adding a prevalence filter, for example keeping only features/taxa detected in at least 2 or 3 samples?
For low-biomass gastric biopsies, should I mainly interpret taxonomy as presence/absence or prevalence rather than relative abundance?
Would genus-level interpretation be more appropriate than species-level interpretation?
I dont work much with oxford nanopore, but two things about your pipeline struck me. First, you've lost a lot of data. You went from 23,784 to 6,258 sequences somewhere along the line. I suggest figuring out where you're losing everything. I'd probably look at that chimera removal step and then taxonomic filtering parameters.
I would also be concerned about your rarefaction depth. I think that can be resolved once you get the first part sorted.
So, I think figure that part out before worrying about the structure fo the data.
I'd like to add that I've also not worked with nanopore data very much.
I'd recommend investigating your uchime-denovo step. I've had problems in the past where many legitimate reads, were inadvertently classified as chimeras, and discarded.
If you are using --p-method uchime and not uchime2 or uchime3, then I'd suggest increasing the --p-minh parameter to a higher value than the default. This can reduce the incidence of what is considered a chimera. This parameter is ignored for uchime2 or uchime3.
I'd try uchime2 and/or uchime3 and compare those results to uchime with varying --p-minh values. At least this is were I'd start investigating your read loss.
In addition to @SoilRotifer and @jwdebelius comments, I would like to add that 97% OTU clustering of 16S nanopore data is not helping much due to the high error rate of the raw reads.
For example, when I clustered 1M of nanopore reads at 97% threshold, I got 900K unique OTUs. In combination with
It resulted in data loss because many copies of the same sequence were clustered as unique OTUs with a count of 1.
This is pretty low. I suspect this will cause issues when calculating diversity metrics compared to 'typical' 16S results using short and abundant (e.g. Illumina) reads.