Making insect classifier

Technical Support
1

Hi, I am trying to make an insect CO1 classifier however it is taking a very long time to do! I am currently using the bold_derep1_seqs.qza I got from the Building a COI database from BOLD references tutorial. I then ran this code to make the classifier specific for my primers:

qiime feature-classifier extract-reads \
--i-sequences bold_derep1_seqs.qza
--p-f-primer GGWACWGGWTGAACWGTWTAYCCYCC
--p-r-primer TANACYTCNGGRTGNCCRAARAAYCA
--p-n-jobs 2
--p-read-orientation 'forward'
--o-reads sequencesampliconspecific.qza

However, as mentioned, it is taking a while to do so. I understand the database is large but I am getting worried that it is not doing anything / I have done something wrong. I have been running it for over 24 hours now. Is there an easier way to get / make a CO1 classifier? Thanks!

2

Hi @graceb,

It can certainly take a long while to build a classifier, sometimes a few days. Do you know how many sequences you have? Have you performed any QA/QC like dereplication?

3

Hi,

Thank you so much for your response! The sequences had these things done to them before I downloaded them:

"The raw BOLD sequences were initially filtered for ambiguous nucleotide content (5 or more N 's), long homopolymer runs (12 or more), very short (< 250 bp) or very long (> 1600 bp) sequences, and dereplicated."

Therefore, I was going to make it specific for my primers, then dereplicate again and build the classifier.

I am not sure how many sequences there are, how would I be able to tell?

Again thanks for your help!

4

Hi @graceb,

Great!

That is what I often do myself. :+1:

You can simply run the following to tabulate your sequences and see how many there are:

qiime feature-table tabulate-seqs \
    --i-data sequencesampliconspecific.qza \
    --o-visualization sequencesampliconspecific.qzv
5

Amazing thanks. For the last bit I tried to run the code and visualise the output on qiime2 view however there is nothing on the screen?

6

Depending on how many sequences are in the file... it can take a while to load as a qzv.

Perhaps try this instead:

# export FASTA
qiime tools export \
    --input-path sequencesampliconspecific.qza \
    --output-path sequencesampliconspecific_export

# count number of sequences
grep -c '^>'  sequencesampliconspecific_export/dna-sequences.fasta
7

That worked great! There are 1718762 sequences - which definitely sounds like a lot!

8

Holy smokes! :scream:

Are these dereplicated? If not try that. Otherwise, when dereplicating try setting --p-perc-identity 99 to perform some minor clustering. Might even have to try 98%... as even if you can construct the classifier it might take a lot of RAM to use it...

9

Ok thanks I will give that a go!

Is there another, easier, way to make an insect CO1 classifier? I tried to use a pre-made one but unfortunately it didn't work with the version of qiime2 I am using...

10

If you have access to the sequence and taxonomy files, you should be able to build the classifier yourself for your version of QIIME 2. Off the top of my head, I am unaware of any exiting pre-compiled files.

Also, it is quite okay to cluster the sequences if needed. There is a measure of practicality to constructing databases. Also, you can try theclassify-consensus-vsearch and classify-consensus-blast if the classify-sklearn becomes untenable.

11

Ok thank you so much! I have dereplicated the sequences to an identity of 98% and managed to almost reduce the number of sequences a lot (now 448211!). Hopefully this will be quicker when trimming according to my primers. Thanks again for your help!

1 Like
12

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.