I have used qiime feature-classifier classify-sklearn to classify my reads with the pre trained classifier silvia-138-99-classifier-nb.qza found here (I used pre-trained as I had some problem training my own classifier similar to those discussed here.
I am assigning taxonomy to my ASVs after merging all 18 batches. However due to some issue with a lack of shared features across samples, discussed here, I have later clustered my ASVs at 100 and 99%. Is it possible to use this previous taxonamy.qza on the later clustered feature tables?
I ask as feature names are all done by hashes of the sequences so i assume after clustering those hashes might change, or are just the centroid features kept and thus those hashes will still match those in the taxonamy.qza?
Yep. This should be the case. However, this can cause problems because you may end up using a very specific taxonomy to represent a cluster of sequences which may vary at the family and genus level. If you'd like to take into account the potentially varying taxonomy of sequences within a cluster, try using some of the RESCRIPt functionality to cluster your sequences and adjust the taxonomy accordingly.
For example try using the command: qiime rescript dereplicate ... Specifically, read through the help text on --p-mode, where you get to decide how to adjust the taxonomy for the resulting representative sequences of a cluster as defined by --p-perc-identity.