I am trying to combine two microbiome 16S rRNA studies both including a disease and controls - study design is similar, trimmed to overlapping reads. After Mothur processing, reads depth of study A is much greater than study B. Before correcting study effect with ComBat, I removed all OTUs (AKA species) only unique to study A or study B (i.e of OTU001 does not appear in study B it is filtered out. This should improve ComBat correction. HOWEVER, I lose all disease samples in study B (7 samples), the study with lower depth - these 7 samples contain 7 or less OTUs after filtering. In fact, these 7 disease samples in that study cluster separately in NMDS plot after normalization (blue triangles on the right) before filtering out those study unique OTUs (similar results without normalization or with rarefaction). I assume that these two studies cannot be combined.... What do you think? Is it still reasonable to combine? Also, no matter what method I try (normalization, filtering, rarefication, study correction) the study A and study B always cluster separately. Any suggestions to fix this problem?