Retraction of Silva 515F/806R taxonomy classifier


(Greg Caporaso) #1

We have discovered an issue with the recently posted Silva 119 Naive Bayes 515F/806R taxonomy classifier. Due to an error in pre-processing, the trimming of the reference sequences to the 515F/806R region was performed incorrectly and in many cases the trimmed sequences only partially overlap this region of the gene sequence.

The classifier is not expected to produce incorrect results, but rather will not perform as well as it should, or as well as the Silva 119 Naive Bayes full-length taxonomic classifier.

The UUID of the affected artifact is 208b55a4-05e2-412b-95b4-a3c7404094d6. You should check the UUID of your classifier artifact to determine if you are using this artifact. This can be done by viewing the artifact at, or using the following QIIME 2 command on the command line interface:

qiime tools peek silva-119-99-515-806-nb-classifier.qza

(Note that the filename silva-119-99-515-806-nb-classifier.qza can not reliably indicate if you’re using the affected artifact. The only way to be sure is by checking its UUID.)

We are in the process of rebuilding the Silva 119 Naive Bayes 515F/806R taxonomy classifier and will post an announcement when this artifact is ready. In the meantime, we recommend using the Silva 119 Naive Bayes full-length taxonomic classifier. Any analyses that were run with the retracted Silva 119 Naive Bayes 515F/806R taxonomy classifier should be re-run with the Silva 119 Naive Bayes full-length taxonomic classifier.

Sorry for the inconvenience!

(Gregory Farrant) #2

Hi hi !
Any update on the moment the classifiers will be available ?
Also I noticed that the OceanSamplingDay consortium decided to start using a new set of primers based on Parada et al., 2016…
Maybe we should look in to this…


(Greg Caporaso) #3

Hi @Gregory,
Sorry for the delay on this. We’re preparing a new release that should be ready within the next two weeks. I’m going to retrain this classifier once that version is ready, and will follow up where when I do. Thanks for your patience!

I’d also be open to training with other primers at that time - if you can post the forward and reverse primers I’ll train with those as well.

(Greg Caporaso) #4

Hi @Gregory, @thermokarst reminded me that I never followed up here. The new classifiers are now posted on our Data Resources page. Let us know if you have any questions. If you’re interested in training classifiers based on other primer pairs, you should refer to the Training Feature Classifiers tutorial. In QIIME 2017.2, the performance of these classifiers was improved so that it’s now feasible to train without huge amounts of RAM (the Silva full-length classifier on the Data Resources page was trained in about 50 minutes - I don’t know the peak RAM usage, but I’m guessing it wasn’t much more than 16GB).

(Gregory Farrant) #5

Hi @gregcaporaso !
Many thanks for the news.
I’ll download the data and make them available to ISCaR users =)

I’ll look through the tutorial since it seems we are actually using various kinds of primers.