Different number of significant DA taxa when changing reference

[AbelSILLY]

Hi,

I'm currently using ANCOM-BC and try to test for difference of composition in a variable with more than two groups.
I use the global test global = TRUE in the ancombc function but it seems like I get a different number of significantly differentially abundant taxa when I change the reference group of my variable. It seems weird to me because I believe this doesn't happen with regular linear models.
Is this a specificity of ANCOM-BC and if yes, is there a proper way to choose a reference ?

Thank you,
Abel