SILVA 138 assigning incorrect species

Hi,

I'm using the pre-trained Naive Bayes classifier trained on the Silva 138 99% OTUs from 515F/806R region of sequences for my data (on Qiime2 v. 2021.2), and have noticed an extremely unusual species assignment, that I cannot explain.

Assignment down to genus seems as usual, but then species is determined as a nematode:

I've examined the taxonomy file closer and found another instance where species is a coral (assigned to g__Endozoicomonas;s__Acropora_cervicornis). And a few other where genus and species are mismatched, but perhaps at least some of those can be explained by the changing taxonomy of gut microbiota:
mistmatched

I have tried re-downloading the SILVA database but that did not change anything. Perhaps training it on my data would be an option, however, I'm not sure how anything like this can occur, the only connection I have found is that the Endozoicomonas bacteria can be symbionts of corals, however, why that would result in the coral being assigned as species for an ASV is a mystery.

Thanks,
Ilze

Hi @Ilze ,
This is a good question — and not unexpected!

Long story short, SILVA does not actually curate the species-level taxonomy. It curates genus level, and provides the species annotation of the "source organism" given for the sequence in NCBI. In many cases, the "source organism" listed is the host of the organism, not the organism that the DNA belongs to :grin:

QIIME 2 (and RESCRIPt, the software used to make the Q2-compatible database) does not do anything to alter this — it takes the taxonomy that it is given, directly from SILVA.

You can check this directly on the SILVA website to see that this mismatch exists right at the source (e.g., g__Endozoicomonas;s__Acropora_cervicornis).

This is discussed a little more in this tutorial (along with steps for building and filtering the taxonomy yourself, including ways to build a genus-level taxonomy if you want to drop the species labels altogether), as well as the following publication:

5 Likes

Thanks for the explanation, I'm sure it will be useful to others using SILVA as well!