@balford looks like your table has a ton of singletons and doubletons, which is known to be problematic whenever you run compositional methods (including ANCOM). Could you try to filter out features that have less than 10 counts in the total experiment? If an ASV only appears 1 time, its likely garbage.