Hi @Emily_Klann - I think your model is correct. If you are concerned about zero-inflation, you can try out one of CPLM, ZICP, or ZINB as your analysis method (we recently included the full functionality of these non-LM models in the most upgraded version submitted to Bioconductor and available in GitHub: https://github.com/biobakery/Maaslin2). Another alternative is to try out strict filtering to remove some of these highly zero-inflated features before running a linear model. The fact that your DESeq2 results are different from a default MaAsLin2 run with random effects is not surprising given that DESeq2 is not capable of fitting random effects and is not appropriate for repeated measures.