ESV vs. OTU workflow

I am getting so confused with the differences between these two and how the work flow would actually change.

I was working with an older script and modifying it to fit my 16S data (515f-804r) with the recently updated Silva database (my old classifier was from the 2015 version). However I’m having a lot of trouble determining if my work flow still will analyze and work for ESV or whether it only will do OTU.

I trained my classifier and trimmed my data and have been using dada2. And then chose a sampling depth that didn’t remove too much of my data. Would this be suitable for esv analysis?

From my understanding, OTU classifies sequences into groups based on a certain percent of similarity (i.e. 97%) but ESV instead keeps everything as independent ‘groups’ (so how would this be done in the workflow?)

Any advice on modifying a OTU workflow to instead be used for ESV is greatly needed.

Hi @Ellenphant!

See this tutorial (check out the flowchart) and this flowchart too.

OTU clustering:

  1. Use quality-filter to filter sequence data.
  2. Use vsearch dereplicate-sequences to dereplicate
  3. Use one of the vsearch OTU clustering methods


  1. Just use the appropriate denoising method for your data in place of the OTU clustering commands above.

The inputs and outputs to/from these steps would be identical, as shown in the flowcharts.

Sounds like you are already doing ESV analysis with dada2. So yes, this sounds perfect.

OTU picking clusters sequences into groups based on a certain percent of similarity (i.e. 97%). Denoising methods instead denoise the sequences to remove noise and dereplicate identical sequences into unique sequence — essentially 100% OTUs with less sequencing noise.

I hope that helps!

Sorry just another quick question and clarification.

So the workflow for otu clustering and esv analysis is still the same in qiime2?

For ESV analysis after denoising my sequences is that it and it’s ready to go? Since ESV don’t need to be taxonomically clustered so there’s no step for that?

Hey there @Ellenphant!

As compared to what?


You might consider some filtering or quality control by filtering your resulting feature table or sequences, but, that is optional.

Keep us posted! :qiime2: :t_rex:

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