If I got it right, Maaslin2 builds a model per each microbial feature, independently of other microbial features (that is also why BH correction can be applied). But I notice that for the SAME bacterial feature, different coefficients are derived, if using a smaller input table with fewer features (subsetting only features, not samples). Taking a simple model as an example, and showing the coeffcieints for Escherichia:
Maaslin2(dat2,key2,min_abundance=0.005,fixed_effects = c(“Diagnosis”),paste0(path,“temp”))
coef for Escherichia: -0.0296
Maaslin2(minidat2,key2,min_abundance=0.005,fixed_effects = c(“Diagnosis”),paste0(path,“temp1”))
coef for Escherichia: -0.041
std. error and p value changes as well. Note that there is no change in the input data for Escherichia itself, only in teh number of features supplied along with it.
And a question perhaps related - all the coefficients reported by Maaslin2 are very different from those reported from glm (in this case) or lmer (for random effects models). Why? Example , again on Escherichia (log10 transformed):
gives: -2.5237 as coeffcient