No taxonomy assignment with pre-trained classifier

Hi all,
I'm working with 16S data from several different studies. I have a trouble on taxonomy analysis now. As most of them are V3-V4 amplicon but some of them are V1-V2 and V4-V5, I used pre-trained silva classifier with Silva 138 99% OTUs full-length sequences (https://data.qiime2.org/2022.2/common/silva-138-99-nb-classifier.qza) downloaded on Data Resources section of docs.qiime2.org. Below is the command I used:

qiime vsearch cluster-features-closed-reference
--i-sequences merged_rep-seqs.qza
--i-table merged_table.qza
--i-reference-sequences ../db/SILVA/silva-138-99-seqs.qza
--p-perc-identity 0.97
--p-threads 16
--o-clustered-table otu_clustering_outcome/table-cr-97.qza
--o-clustered-sequences otu_clustering_outcome/rep-seqs-cr-97.qza
--o-unmatched-sequences otu_clustering_outcome/unmatched-seqs-cr-97.qza

qiime feature-classifier classify-sklearn
--i-reads otu_clustering_outcome/rep-seqs-cr-97.qza
--i-classifier silva-138-99-SSU-classifier.qza
--p-n-jobs 16
--output-dir taxonomy_outcome

After doing them and get taxa barplot with qiime taxa barplot command, I found out that 2 out of studies I'm working on are almost not assigned like below:

The studies almost not taxonomical assigned are V3-V4 amplified data, but the thing is other studies with same amplified region are well assigned (even the studies of V1-V2 and V4-V5 are also well assigned). I double checked all of samples have enough reads after denoise process (all more than 700 reads).

Any idea what was the problem? or any ways to get beautiful assignment?

Hi @Jonathan,

Thanks for reaching out!

I checked in with another moderator on this, and it seems highly unusual. Is there any possibility that you clicked/un-clicked the boxes on the right to reveal/hide some groups?

Otherwise, since you used closed-reference OTU picking, there should be taxonomy assignments. If the features are retained via closed reference, then that is a good sign these are real sequences and would have an assignment.

Do you mind sending us your .qzv file so we can look through the provenance?

Thanks! :lizard:

Hi @lizgehret ! Thank you for your reply.

Nop, the picture on the post was the very first view when I open the qzv file.

I also tried using customised trained classifier, but it also gave me the same assignment (I trained 3 different classifier with 3 different region; V1-V2, V3-V4, and V4-V5 referring to primer sequences written on their papers).

And I've sent you a message with the tables from denoise (dada2 single) and closed-reference clustering :slight_smile: