Do I get it right that Maaslin3 should NOT be used to study effects OF microbiota ON something (like, growth of an animal)? I mean, microbiota features as exposure and growth as outcome? Why is this wrong, if e.g. significant linear associations are observed? May I ask what would be an appropriate method for such cases?
This is a can of worms, but in general, you can either look at taxa ~ covariate(s) or covariate(s) ~ taxa. Typically, the answers will be similar. MaAsLin answers taxa ~ covariate(s), which gives the interpretation “the species changes by Y fold for every X unit change in the covariate(s).” If you’re trying to answer covariate(s) ~ taxa, you’re in the world of high-dimensional inference, machine learning, or whatever else you want to use to explain a low dimensional outcome from a large set of predictors.
1 Like