Imprecise taxon classification...but only in one group!

[Nicholas_Bokulich]

Hi @skaup ,

Welcome to the forum! I think I know what is going wrong. This sounds quite similar to this topic:

Taxonomy assignment with custom COI database - odd/wrong classifications?

Hello all, I am running a custom COI database from BOLD that took a few days on my HPC to extract reads and train, following the instructions on the qiime2 docs. When I run feature-classifier classify-sklearn on my data (marine water samples), almost all of the sequences come back as unidentified Arthropods, or oddly, birds. I am a bit stuck here, as I tried this custom database to compare with Midori (and the same steps with Midori seemed to give back reasonable results - though left more to b…

The issue is most likely that the database you built contains one or more of:

1. very short sequences that skew classification
2. misannotated sequences

How did you build your classifier? You can use the plugin RESCRIPt to visualize the length distribution and filter your sequences by length to remove any outliers that are causing possible issue #1. See also the topic I shared above (and the posts linked therein) for some more troubleshooting.

@skaup

In which case, why only bony fishes?

The issue is probably not with bony fishes per se, but that there are 1 or more reference sequences that are really short and/or misannotated as bony fishes. Hence you have a very high number of query sequences being classified as Actinopterygii. So the ASVs are either being misclassified as Actinopterygii, or matching a reference sequence that is misclassified as Actinopterygii.

I suggest filtering/looking at the length distribution first to see if this already improves your results... if not, you should see if there are any reference sequences that are only classified at the class level.