I have a question about the QIIME assign_taxonomy.py script with the RDP classifier.
When I run the assign_taxonomy.py script several times on the same input data, I get slightly different results. For example, sometimes an OTU is assigned up to genus level, other times the same ID sequence is assigned up to family level only.
Is it possible to set a seed, so that to get the exact same results when running the script several times on the same input data?
Also, will the RDP classifier be implemented in QIIME2?
Dear @dbau,
Welcome in the forum!
This question may probably be better located in “Other tools” given rdp is not in qiime2 (as long as I know … if not sorry!)
On the question, for what I remember, rdp do a on-the-fly-training on the database and that probably why you have slightly different results. I am not aware of a way to keep it fixed between runs.
Have you tried to assign taxonomy within qiime2? That would split the training step form the assignment step, so if i am right it may result in what you need!
I am not sure if they will implement rdp within qiime2, they may rely on external developers to provide a third party plug-in.
We do not have any plans — check out the feature-classifier plugin, which has two methods (fit-classifier-naive-bayes and classify-sklearn) that train and use a naive Bayes classifier to assign taxonomy to your sequences. This is the same underlying method used by RDP, and the two have very similar performance.
And @llenzi is exactly right: we would be happy to have RDP join the QIIME 2 family, if any developers out there want to write their own plugin to wrap RDP.