You can take your denoised reads (ASVs) and run closed-ref clustering on them to make closed-ref OTUs. That was done on the American Gut project that Justine linked, and it may make sense in some contexts because it gives you both 1) new-school ASVs and also 2) old-school database OTUs.
In that thread, the user asks if doing de novo clustering then dropping NAs is the same as closed-ref, and no, no it's not.
The options are
denoising -> ASVs (this is the modern, recommended option )
de novo clustering -> de novo OTUs
closed-ref counting -> database OTUs
closed-ref counting followed by de novo clustering -> open-reference OTUs
denoising followed by closed-ref counting -> database OTUs (from your ASVs)
(this is the hybrid approach you asked about!)
de novo clustering followed by dropping NAs -> this is nothing
Many thanks for your clarification, really helpful.
I was wondering, do you think the closed-ref counting on my ASVs is different from the closed-ref counting approach? or just a lesser number of features?
Yes. But let's start with their major similarity: both methods will match against a selected database, and return the features from that database that match your inputs to a given % identity. The resulting features will be 100% biased by == 100% consistent with that database, and this adherence to the database is needed for some methods.
Denoising before doing this database search will give you ASVs as features you can use for other pipelines or database-independent methods.