Testing for differential abundance and diversity for different host traits

General Discussion
1

Hey,
I’ve used 16s to study the diversity and composition of the microbiome of dozens of mammals.
I want to find differentialy abundant ASVs between groups of hosts based on host categorical traits (such as social structure) while taking into account the phylogenetic confounding factors.

I’ve found [phylogenetic ANOVA] but ANOVA assumes normal distribution which is not the case in abundance or relative abundance of bacteria.

Do you have any suggestion of non-parametric test that accounts for the host phylogeny?
(https://www.rdocumentation.org/packages/geiger/versions/2.0.6.4/topics/aov.phylo)

2

I’m hoping to do some differential abundance analyses myself, and have been reading some papers. This review paper suggests a couple of methods for compositional data that I plan to work with. Hope it helps a little.

1 Like
3

Hi @Jes_biol,

I haven’t worked much in comparative phylogeny, but there was a recent paper that addressed it across a wide variety of species. Maybe that methods section could help you?

One thought (off the top of my head, feel free to disregard) is to look at a relationship between your host phylogenetic distance and your community disimilarity using somehting like a procrustes analysis. For this, I can imagine turning your host tree into a distance matrix (I think you can do this in scikit-bio and then import into QIIME 2 using the artifact API) and then looking at the relationship between that and your community measured with your favorite metric(s).

I think this might actually be a really interesting place to apply some of the tree based methods. I know gneiss is being slowly decomissioned and ILR interepretation is a headache and a half, but you might be able to do some kind of clustering between the organism and the phylogenetic distance?

Best,
Justine

2 Likes
4

Thanks for referring me to compositional methods, however my issue is with treating phylogenetic confoundments

1 Like