Vsearch Concensus is giving lots of unassigned reads

[victoriamesa (victoria)]

Hi @jwdebelius,

Excuse me again, I continue having some problems.
Finally I chose only to work with forward sequences due to low quality scores in the reverse.
My barplot looks as follows, with a lot of unassigned sequences in some samples.

Here, my scripts:
qiime cutadapt demux-single
--i-seqs MULTIPLEXED_SINGLE/multiplexed-seqs.qza
--m-barcodes-file MULTIPLEXED_SINGLE/metadata.tsv
--m-barcodes-column Barcode
--p-error-rate 0
--o-per-sample-sequences MULTIPLEXED_SINGLE/demultiplexed-seqs.qza
--o-untrimmed-sequences MULTIPLEXED_SINGLE/untrimmed.qza
--verbose

It seems strange to me the low percentage that indicates with adapters?

qiime cutadapt trim-single
--i-demultiplexed-sequences MULTIPLEXED_SINGLE/demultiplexed-seqs.qza
--p-front GTGCCAGCMGCCGCGGTAA
--p-front GGACTACHVGGGTWTCTAAT
--p-error-rate 0
--o-trimmed-sequences MULTIPLEXED_SINGLE/trimmed-seqs3.qza
--verbose

qiime dada2 denoise-single
--i-demultiplexed-seqs Gut_microbiota/MicrobiotaPerro/MULTIPLEXED_SINGLE/trimmed-seqs3.qza
--p-trim-left 20
--p-trunc-len 220
--p-n-threads 2
--o-table MULTIPLEXED_SINGLE/table3FORW.qza
--o-representative-sequences MULTIPLEXED_SINGLE/rep-seqs3FORW.qza
--o-denoising-stats MULTIPLEXED_SINGLE/denoising-stats3FORW.qza
--verbose

denoising-stats3FORW.qza :

qiime feature-classifier classify-consensus-vsearch
--i-query MULTIPLEXED_SINGLE/rep-seqs3FORW.qza
--i-reference-reads ref-seqs97.qza
--i-reference-taxonomy ref-taxonomy- silva_132_97_16S.qza
--o-classification MULTIPLEXED_SINGLE/taxonomy-vsearch.qza

In taxonomy-vsearch I don't see many unassigned:

I appreciate your review and guidance, I don't know how to continue.
Thank you so much

[Nicholas_Bokulich]

Hi @victoriamesa,
There has been a lot of good discussion on this forum about the sources and causes of unclassified sequences, I suggest starting here to diagnose:

Train you classifier - taxonomy

Hi @Jasmine84, See this post. Depending on the length of your sequences, the primers you are using, etc, deeper than order level might be unreliable. On 250 bp 16S rRNA gene sequences, we can usually get genus level and some species with a good degree of reliability. The naive bayes classifier in QIIME2 is designed to be cautious and reliable, to avoid false-positive errors, but you can lower the confidence parameter to make it less cautious. If you have: only kingdom-level classification, …

In your case, I also notice that some of your samples have very few sequences... at least a few of these appear to be the samples with very few classified reads. So those are samples that should probably be dropped anyway, and the reads in those samples could just represent some background noise (e.g., reagent contamination, index hopping?).

Good luck!