Classifier outdated?

[Paul Fuhrmann]

Hello, I wanted to use the MIMt DB to align to my microbiome dataset (I got it pre-aligned to SIVLA and RDP but was not happy with the taxonomy provided). I tried doing it myself with the current 16S MIMt classifier via QIIME2 and got this message:
ValueError: The scikit-learn version (0.24.1) used to generate this artifact does not match the current version of scikit-learn installed (1.4.2). Please retrain your classifier for your current deployment to prevent data-corruption errors.

How should I proceed? Can I force the command or is it safer to downgrade scikit-learn? 
I tried training the classifier myself but my hardware did not enjoy the process and killed the command. I apologize for my lack of knowledge, I am new to this.   

Kind regards, 

Paul Fuhrmann 

[Arval Viji Elango]

Hi Paul,

To preface: I am not associated with the MIMt team, nor have I used it before. I just learnt about MIMt from the QIIME2 Forum, and came over to have a look. 

Having said that, the issue you face is not with MIMt but with the QIIME2 version used. QIIME2 updated a bunch of packages it is dependent on in some of its recent updates (namely v2024.5). If you are using a newer version of QIIME2 and receiving this error, it means the MIMt classifier you are using was trained on an older version of QIIME2. 

Solutions to this: Use the classifier with an older version of QIIME2 (pre-2024.5) to classify your data. Or train your own classifier (which, as you have found out, is very computationally intensive).

Thanks,

Arval

[Antonio Muñoz Mérida]

Hi Paul and Arval,

said that, I can add that a new version of MIMt has been released on the weekend and all the classifiers for all the markers were trained with the last version of QIIME2, so now instead of training the classifier by your own you can use the new versions that for sure they should work on your data.

Thanks Arval for clarifying the problem with the QIIME2 versions

Antonio