please help creating human-stool-weighted for v3v4 silva (q2-clawback)

Library Plugin Support
1

Hi,

So I’ve been trying to train silva classifier to fit my v3-v4 samples. And since my sample is human stool sample, I thought it would be good to train it into human-stool-weighted.

So what I did is extracted v3-v4 region using below command:

qiime feature-classifier extract-reads
--verbose
--i-sequences silva-138-99-seqs.qza
--p-f-primer CCTACGGGNGGCWGCAG
--p-r-primer GACTACHVGGGTATCTAATCC
--p-trunc-len 0
--o-reads silva-138-v3v4-seqs.qza

and then created uniform v3-v4 classifier

qiime feature-classifier fit-classifier-naive-bayes
--i-reference-reads silva-138-99-ref-seqs.qza
--i-reference-taxonomy silva-138-99-tax.qza
--o-classifier silva-138-v3v4-tax.qza

since I can’t find human stool sample in qiita, I tried to directly class-weighted it with human-stool.qza downloaded from https://github.com/BenKaehler/readytowear/blob/master/data/silva_138_1/full_length/human-stool

with command:

qiime feature-classifier fit-classifier-naive-bayes
--i-reference-reads silva-138-v3v4-seqs.qza
--i-reference-taxonomy silva-138-99-tax.qza
--i-class-weight human-stool.qza \
--o-classifier silva138-V3V4-human-stool-classifier.qza

but it shows Plugin error from feature-classifier: Number of priors must match number of classes.

Is there anyway I can bypass this error? Also if there is any advice on how to use the qiita for human-stool, I would really appreciate the help.

Regards,

Natasia

2

Hi @futoncrawler ,

This error indicates that your sequences and/or taxonomy do not match the weights. Either (1) you are using different versions of the SILVA database or much more likely (2) when you use extract-reads it is dropping sequences that fail to trim due to mismatch with the primers. In the case of #2, the solution is to filter your taxonomy to match the sequences — this can be done using RESCRIPt, which has a filter-taxa action (you should be able to pass the sequences artifact as --m-ids-to-keep-file) — and filter the class weights to drop taxa that are missing from the taxonomy (this could probably be done with qiime feature-table filter-features but I have never tried).

If that does not work the best option will be to just re-create the class weights following the tutorial.

Good luck!

1 Like