I have seen peer-reviewed publications where Deseq2 is used for differential abundance analyses of 16s rRNA data. I've recently been informed that Deseq2 may not be appropriate for analysis of 16s reads. Can anyone provide the evidence-based sources that explain why Deseq2 should not be used for microbiome data and provide any alternatives to hypothesis testing using 16s data in R?
I think the big assumption where DESeq2 would fall short are that there are features that are relatively constant across all samples in true abundance, and that we can find these features using our typical methods (median ratio size factor estimation). However, it may also be that the conditional distribution is not Gamma-Poisson (NB), but something more like LogNormal-Poisson.
I know some of the alternative methods in this space are ALDEx2 and corncob.