Rarify before NMIT?

I have a simple question about the nonparametric microbial interdependence test (NMIT). Should I rarify the FeatureTable beforehand?

NMIT is operating on relative frequencies, so you probably don't need to rarefy, unless if there is a big difference in sequencing depth between sample types/groups/times/whatever. But I am not aware of the statistical assumptions that might be violated one way or the other by (not) rarefying.

The author of this method, @Yilong_Zhang , may be able to tell us more.