Hi! Is it always frowned upon to use different qiime2 versions throughout your workflow? I'll put a couple examples below.
Most of my workflow was under the latest version of Qiime2 (v 2026.7). However, colinbrislawn's latest pre-trained UNITE classifier is trained for version 2026.4. So, I input my table and data artifacts generated using v 2026.7 to classify-sklearn, metadata tabulate, feature-table tabulate-seqs, and taxa barplot under v 2026.4 just to assign taxonomy to my data. I then moved back to v 2026.7 for the rest of my workflow: DADA2 v 2026.7 Sklearn v 2026.4 Decontam v 2026.7
At first, I thought that ITSxpress only worked under v 2024.10, so I attempted to do the same as above but only for that filtering step: Imported data v 2026.7 v 2024.10 ITSxpress v 2026.7 DADA2(It's possible this choice might have actually caused an error where Qiime wouldn't let me proceed due to mismatched versions.)
Most Qiime output formats are forward compatible: you can use qiime-2026.7 to open a file made with qiime-2023.7.
The sci-kit learn classifiers are a little different, as we pretrain them and them ship the binary classifier, so the version numbers for skl need to match exactly. Note that improvements to the q2-feature-classifier plugin could make the SKL databases forwards compatible too!
In this specific case, I can confirm that you can use QIime 2026.7 with the pre-trained classifier I made with 2026.4 because they both use scikit-learn=1.7.1.
That being said, I like using the same version of Qiime2 for a full project! Keeps things simple and you don't have to justify this to reviewer three.
What happens when you assign taxonomy using qiime-2026.7 and the 2026.4 unite-train classifiers?
Hi Colin, thank you for your reply! I didn't even try to use your classifier trained under 2026.4 in qiime-2026.7... but I'll go ahead and do that now.
Sklearn v 2026.4 
