Picrust2 accuracy / internal validity

I am developing a tool that takes as input the presence/absence of KO pathways predicted by picrust2. Since the errors made in prediction will propagate through to the final result I have been looking for a summary of the internal validity of Picrust2’s predictions on a per-pathway basis, i.e. leave-one-out cross validation on the database used in the manuscript. Has this analysis been done and is it available somewhere?


The code for that manuscript appears to be in this GitHub repo:

I have not tested this myself, but would be interested in your experience with it (and your new tool!)


P.S. Welcome to the forums! :wave:

1 Like