Hello everyone. I know that differential abundance analysis is still an active research area, that there exists multiple methods (ANCOM, ALDEX2, DESeq2...) to detect significant ASV/OTUs, and that there is no gold standard yet, each method having its drawback (distributional assumption, pseudocount, lower sensitivity on small dataset etc...) as stated here.
However, my question concerns differences at higher taxonomic ranks (e.g. Phylum, Class...). In that case, what is the best way to proceed, statistically speaking ?
From my understanding (I am not a statistician ;)), we can't use DESeq2 or ALDEX2 because they make distributional assumptions at the ASV/OTU level that may not be appropriate at higher ranks. ANCOM, making no distributional assumption, and working on feature ratios seemed legit to me, but my dataset is rather small and it has been shown that its sensitivity is quite low on small dataset.
So now, I am considering a simple TSS normalization followed by either ANOVA (if features are normally distributed) or Kruskal-Wallis test.
Do you think this approach is OK ? Is there a more appropriate normalization/test combination (e.g. CLR + KW) ?