Hi
I have 30 samples (at 3 different timepoints) for which I have both metabolome and microbiome data available. I am doing following analysis (after pre-processing):
- bacterial abundances and scaled metabolite abundances were rank normal transformed using the rntransform() function taken from the GenABEL package and regressed against a set of confounders.
- Using the residuals as new metabolic/microbial traits, individual linear mixed models were applied between all metabolite-genus pairs using the lme4 R package, to find significant associations between individual metabolites and bacterial genera. P-values were obtained using F-tests and were corrected for multiple testing. Significant associations with a FDR cutoff q<0.01 were retained.
We are interested in how the presence of metabolites might perturb microbial composition as well as in the other way around. I wonder know whether it does make sense to analyze metabolite and microbe abundances are both analyzed as dependent variables , as follows:
a) Associations of metabolites with microbial covariates
metabolite ~ microbe (ASV) + month + (1|subject)
each of the metabolites is regressed against each of the ASVs.
b) Associations of microbes with metabolite covariates
Microbe (ASV) ~ metabolite + month + (1|subject)
each of the ASVs is regressed against each of the metabolites.
I did, and I found ~900 significant associations in (a) whereas only 250 in (b), can we conclude that metabolites are more significant in explaining the microbial abundance then? I am note sure about this statistical insight and I wondered whether there are other ways to do so?
Any help would be appreciated,
Thanks in advance!
