pre-trained UNITE ITS classifiers for QIIME 2026.1 (and older!)

Community Contributions Data resources
, , , ,
1

UNITE version 10.0 was updated on 2025-02-19, and I've built classifiers for it compatible with the newest version of Qiime2:

New in 2026

  • I've rewritten the pipeline in Nextflow, instead of Snakemake. :green_apple:
  • I now use RESCRIPt to get the data, edit the taxonomy, and collate reports! :memo:
  • I've added a simple reclassification step to measure best-case performance. :bar_chart:
    • this shows how the singleton classifiers perform slightly worse :chart_decreasing:
    • and using all Eukes performs slightly better! :chart_increasing:

Not what you are looking for? Please send me a direct message or open an issue.

Colin :mushroom: :bird:

37 Likes
UNITE+INSDC Dataset for ITS fungi
AMF classifier new version
QIIME 2 2022.8 is now available!
how to parse UNITE references for taxonomy assignment with blast?
FUNGuild: Found 0 matching taxonomy records in the database
pretrained classifier for UNITE (ITS OF FUNGI)
How to train a UNITE classifier using RESCRIPt
2

Thank you very much Sir

1 Like
3

I see "all" eukaryotes? Does this mean CO1?

4

Good question! I think UNITE is ITS only (so, no CO1 to my knowledge)

You can read more on their page: UNITE - Resources

5 Likes
5

Hello @colinbrislawn - I have a novice question about using this pre-trained classifier.

The qza files available in the GitHub link you posted, can these be used directly in the "test the classifier" portion of this q2-feature-classifier tutorial?

Ex:
qiime feature-classifier classify-sklearn
--i-classifier unite_ver9_dynamic_25.07.2023-Q2-2023.9.qza
--i-reads rep-seqs.qza
--o-classification taxonomy.qza

1 Like
6

Hi @Taylor_Akers,
From what I understand, this should work!

3 Likes
7

Great, thank you @cherman2 ! It worked!

2 Likes
8

An off-topic reply has been split into a new topic: qiime2 does not recognize classifier as a .qza artifact

Please keep replies on-topic in the future.

9

Hi @colinbrislawn thank you for creating this classifier!

I downloaded the latest version and tried it with my data. However, it runs for about 20 minutes before the job is killed. I am assuming this is due to insufficient RAM (I have 16GB)?

In that case, I should likely create my own classifier using RESCRIPt? Thank you.

2 Likes
10

Making a new classifier takes more memory then using an existing classifier. Using one of these pre-made ones still probably your best option with limited memory...

I would try using a smaller number of reads per batch, say
--p-reads-per-batch 5000 or 1000.

Also consider using a search-based classifier like
qiime feature-classifier classify-consensus-blast

3 Likes
12

I have the same issue—after updating QIIME2 to version 2025.4, the classifiers no longer work. When I try to use them, the execution gets interrupted. I don’t think it’s a RAM issue since I have 16GB, and I had no problems with the previous classifiers and QIIME2 version 2023.9. What could be the cause?

13

Hi @Steven_Criollo, Are you using newly trained classifiers, or classifiers that were trained for QIIME 2 2023.9? The classifiers need to be updated every time we update to a new version of scikit-learn, and we did that in 2024.10, so pre-2024.10 classifiers won't work with 2025.4. @colinbrislawn has new ones trained which you can download from his git repository.

2 Likes
17

2 off-topic replies have been split into a new topic: Can I use the UNITE database for LSU data?

Please keep replies on-topic in the future.

18

I may have re-read this and confused myself, so this does mean that the pre-trained UNITE v10.0 2025-02-18 for qiime2-2024.10 would be compatible with qiime2 2025.4, 2025.7, or 2025.10 until UNITE releases another update or when a new version of scikit-learn is included in the updates? Thank you.

19

Yes! I trained these with skl-1.4.2 so as long as the Qiime2 conda environment uses sci-kit learn version 1.4.2, these will work!

1 Like
20

Great! Thank you for clarifying. Based on the other replies, it did seem that they would be compatible, but I didn't want to assume.

Also, thanks for sharing the pre-trained classifiers. I do have a suggestion for the github. I think it would be great to include in the notes which versions of qiime2 are compatible with which classifiers in order to make it more obvious, especially for those who may get overwhelmed with all the new qiime2 version that get rolled out.

1 Like