[Question] Revisiting Multi-omic Analysis

[mirpie]

Hi there!

First off, thanks for such an awesome package.

Second, I was hoping for some help integrating it into a complex study analysis I'm doing. It involves microbiome and flow cytometry data measured from the same animals in a cross-sectional manner over time (4 independent timepoint groups from 1-8w).

I saw your previous post regarding multi-omics as well as your recent commentary on the validity of proportionality for multi-omic studies. I just want to make sure I'm on the right track conceptually to apply these ideas to my own study.

Ideally, we're trying to examine which microbial populations (50 differentially abundant ASVs from upstream analysis) correlate with immune cell populations measured in the gut, spleen and thymus (each of which constitutes a separate data matrix).

I know this kind of technique is not standard for flow, but I guess you could say I've drunk the CoDA kool-aid a bit and want to demonstrate it's utility for this application. On the microbe side of things, I'm very comfortable with these techniques, and the data structure makes it easy: in this case, I just extract the CLR values from the full transformed data matrix for my taxa of interest. However, for the flow data it's a bit more complicated given the nested nature of the compositions (i.e. every population is measured as a fraction of the parent gate).

My strategy here would be to perform separate CLR transforms for each "level" of resolution: e.g. do all CD4 vs CD8 cells as a fraction of CD45+, as well as a separate transform where I split the CD4+CD8- population into CD25+/- but retain the other populations at that level to maintain the constraint on the whole. I would then combine the values for all my populations of interest that have been transformed using there appropriate denominators into a single DF that I could measure against my microbe data using propr.

Two things: 1) is this sound, given my final dataset will not comprise parts of the same whole but still represent valid ratios when transformed as I described and 2) how can I adapt my interpretation of the results accordingly?

To be clear, I am NOT looking to correlate immune cells with other immune cells (I understand this would not work with the method above) but am just looking at microbe-immune proportionality.

Thanks so much in advance!!
M