Tax bar plots of CR-clustered vs. feature-classified data

Hi everybody,

there was this question that came up in a separate discussion: If I can use the closed reference clustering to obtain abundances for a taxa barplot, what do I need a classifier for?


1 Like

Thanks for asking @steff1088! There are a couple reasons why you would still want to perform taxonomic classification:

  1. If you use closed-reference OTU picking, you do not need a classifier (theoretically) because the taxonomy can be assigned to the nearest match in the reference database. This is, however, the equivalent of taking the top BLAST hit to assign taxonomy, and is highly discouraged! Just because a sequence has a close match in the reference sequences does not mean that it is an exact match, nor does it mean that there are not other similar matches (even other exact matches) in the reference database and hence that might not be the correct taxonomic assignment. A separate taxonomy classifier should still be used to assign taxonomy so that confidence/consensus can be used to prevent overclassification (note that the blast/vsearch classifiers in QIIME 2 take a consensus assignment, not the top hit unless if requested, for this very same reason)
  2. In the related thread you describe using open-reference OTU picking, not closed-reference. The de novo OTUs clustered by that method will not have an associated taxonomy inherited from the reference database, and hence these must be assigned a taxonomy using an appropriate classifier.

I hope that helps!


This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.