Is it valid to interpret rarefaction curves from DADA2 denoised feature table?

[JeremyTournayre]

Hello,

I wonder if we are allowed to interpret rarefaction curves from a DADA2-denoised feature table, since I read that Good's coverage is not valid after DADA2 denoising because singletons are filtered out by default (see the topic below).

Is it valid to calculate Good's Coverage on DADA2 denoised feature table?

I have calculated Good's coverage on feature tables derived from 16S data (rarefied to 33,500 seqs/sample) and ITS data (rarefied to only 970 seqs/sample) and found Good's coverage to be 1.0 for all samples using qiime diversity alpha --p-metric goods_coverage. This struck me as odd given the low rarefied depth for the ITS samples (though these fungal communities are probably less diverse than the bacterial communities). On further investigation, I found (here) some references to certain measu…

Can we still consider samples whose curves reach the asymptote as having sufficient sequencing depth?

Cheers,
Jérémy Tournayre

[ebolyen (Evan Bolyen)]

Hi @JeremyTournayre,

You can of course interpret rarefaction curves in general after DADA2, but I think as the topic points out Good’s coverage should probably not be the curve you use. Doing something simple like observed features, Shannon, and Faith’s PD are all still completely fine.

The other index that comes to mind as being a problem post-DADA2 would be Chao1 which also looks at singletons.

---

Now if you must use one of these singleton-based indices, then there is an option in DADA2 called “pooled” and “pseudo”. These in principle would permit singletons if say an ASV cluster is decided in sample A but is then seen once in sample B. (The cluster in A allows a singleton in B instead of that ASV being sent to a nearby cluster within B.) But I would probably opt for using a non-singleton index, because that situation is probably somewhat rare and I have no description for how much that would or wouldn’t bias a given index.