Hi Will,
Thank you for the reply! I tried the models using TSS and LOG in MaAslin3. The features I want to test are time, treatment and their interactions. I used two different formulas. One is ~ Time + Treatment + Time:Treatment + (1|SubjectID). I have six samples from the same subject. Then I created a new interaction column in my metadata named as Trt_time to manually combine the time and treatment. Then run another model with ~ Trt_time + (1|SubjectID). While the two models have different outputs of species (some overlapped), p and q values in significant results for the interaction. Why? I set the reference level for the model 1 as reference = c(‘Treatment,TrtA’ , ‘Time,baseline’) and model 2 as reference = c(‘Trt_time,TrtA_baseline’).
Another question for the model 2: If I set the reference level: reference = c(‘Trt_time,TrtA_baseline’). Any significant associations in the final output were generated by only comparing others with the reference level, right? Then I created a contrast matrix to compare different treatments at the same time points (e.g., trtA_baseline vs trtB_baseline). In my significant results for the model ~ Trt_time + (1|SubjectID), I has one row/species having qval_joint<0.01 for trtB_baseline, I believe this siginificant result is against my reference ‘trtA_baseline’. While in my contrast test output, all the p and q values in the rows of abundance and prevalence results are NA. Is there any possible error? I created the contrast matrix using ‘name’ column in the model result output.