make qiime2 2026.4 compatible with classifiers made with Sklearn Version 1.4.2?

Library Support Silva
1

Hello, I upgraded my qiime version to 2026.4 and downloaded the latest qza classifiers from SILVA 138.2. When i tried to run the command qiime feature-classifier classify-sklearn i got the error 'The scikit-learn version (1.4.2) used to generate this artifact does not match the current version of scikit-learn installed (1.7.1). Please retrain your classifier for your current deployment to prevent data-corruption errors'.

I have got around the error by clonning the qiime 2026.4 version into a new environment and then forcing the installation of 1.4.2, which i am aware is a 'dirty' solution that can create many incompatibilities. Hence, my question is: is there any other elegant way of running the 'qiime feature-classifier classify-sklearn command' -e.g. a fixy in the command options to run scikit-learn with a specific version, that does not require building the classifiers ourselves but actually using the resources already available?

Just checking that I am not missing something

2

Hi @Carla_Perez_Mon,

The best approach would be to follow the general approach outlined here to construct your own classifier.

You can choose which steps you'd like to keep/skip and the order in which you lke to run them. For example, if I want to make an amplicon specific classifier (you do not have to; but it does help if you do not have enough memory on your machine to construct a classifier) I often follow this approach.

2 Likes
Looking for pre-trained SILVA 138 V4 classifier for QIIME2 2024.10
3

Hi @Carla_Perez_Mon,

We are expecting to have the trained SILVA classifiers soon. I will try to circle back to this thread when they are available.

Thanks!

4 Likes
5

Hello!
I trained the Silva 138.2 (full-length) classifier a week ago for 2026.4. if you want to use it, you can download it here: classifiers.

If you need the sequences and taxonomy files, they're in the DB_files folder. Full-length classifier can be found in the q2_26-4 directory.

8 Likes
6

This is absolutely a nightmare on my side too haha, I do not know how people are retraining their classifiers, but everytime I try to install q2-clawback plugin, I get massive python dependency conflicts with te python version qiime2 2026.4 uses and so I have been stuck on this for about three days now. I coud just be a not very good bioinformatician too, please help with fixing this.

7

Hey @George_kitundu,

q2-clawback hasn't been updated in quite a while and is almost certainly no longer compatible with the latest versions of QIIME 2 (based on the dependency conflicts you're mentioning). I would take a look at a couple of the earlier replies in this thread - you can either use the RESCRIPt tutorial outlined by @SoilRotifer (which is included in the 2026.4 qiime2 distribution) to retrain yourself or use the classifier @timanix trained with 2026.4.

We do expect SILVA to have trained classifiers available for 2026.4 in the near future, but until then I would try these options as intermediaries.

Cheers :lizard:

3 Likes