Differential abundance analysis at high taxonomic ranks

Hi there,

I’m using several methods (i.e. DESeq2, ANCOM, Aldex2) for differential abundance analysis at the ASV level. I am wondering about which of those methods are still suitable (statistically speaking) at higher taxonomic ranks (phylum, class…) when aggregating taxa. Especially, Aldex2 and DESeq2 make distributional assumptions about the ASV counts and I am not sure if it holds true when aggregating taxa.

What would be the most appropriate method or normalization/test association (e.g. CLR + Kruskal-Wallis) to perform such analysis ?

Hi @Erwan,

Welcome to the :qiime2: forum!

I’d argue the most appropriate high-level test is… none. By the time you get to that upper level, my expereince has been the aggregate behavior you’re seeing is usually either driven by a small number of highly differential ASVs/genera. So, you’re sacrificing resolution for convenience may learn less.

If you want a nested approach, I’ve found Phylofactor to work pretty well. It’s not in QIIME2, but it does provide interesting results.

Best,
Justine

1 Like

Thanks Justine,

I’ll have a look at Phylofactor

1 Like