Hi @yokmok,
I think you've got two issues you need to contend with here. One of your major challenges is going to be each person's individual microbial identity. The microbes you start with will influence the microbes you have to respond to treatment and the way they respond. (This is a whole arm of exploration we don't talk as much about.) I think most of your tests are going to be sensitive to sparsity, if not outliers. You might choses to handle this by filtering (i.e. retaining features that present in at least X samples). This will also improve your FDR. I think the log ratio transform may also help. Of course, these methods are also more conservative than LefSe, so they may struggle with a smaller sample size. I'll link to a post talking about some of these methods below.
Best,
Justine