Comparison of spatial ligand receptor interacttions across groups

[rbutleriii]

Hello,

Related to #862 , I am looking at spatial relationships across experimental groups, and I was wondering if there was (1) a best stat to rank a change in...fold change, and (2) if there would be an appropriate statistical test for confidence in this.

Initially I ranked L-R pairs by absolute difference in L2FC, and it seems pretty interesting as we can see a specific mouse disease model effect, and subsequent change with therapeutic compound (on the compound's target receptor).

I also thought about using change in PI as that incorporates confidence, but in terms of measuring PI difference across groups, some things will have low PI for low L2FC, and some for low p.adj even if the L2FC doesn't change.

I am not sure if it helps, but Liana was using tensor based factorization to define the differential effect of interactions across groups, although I think they then just did a t-test of the factor, and then extracted LR loadings on the significant factors. Ultimately, I kind of want to compare my snRNAseq data to my spatial data, so it would be good if they were somewhat relatable.

Thoughts?

P.S. - for that matter the same comparison question arises when comparing icfs, though at this point I am largely just comparing the change in total number of padj significant icfs between groups. Change in ICF effect size is kind of the same basket.