Thanks everyone for your ideas on how to work around this error! I'll try a different chunk size and follow up on how it goes.
@gmdouglas, is there any way to run q2-picrust2 on smaller batches of samples and merge the predicted feature tables? This is related to a previous question I had about inconsistent placement of ASVs shared across samples when I tried this approach. I'm bringing up the question here because it might help with processing large datasets. Is it possible to process samples independently and merge using q2-picrust's full-pipeline command instead of q2-fragment-insertion + custom-tree-pipeline?
Thanks!