Comparing cell-type-level communication scores across samples

[kestrel614]

Thank you for this wonderful tool!

I obtained cell-type-level CCC scores and their associated p-values for multiple samples through the following commands:

pathway = "ACTIVIN"
ct.tl.cluster_communication(adata, database_name = 'cellchat', pathway_name = pathway, clustering = 'cell_type', n_permutations = 100)
ccc_scores = adata.uns['commot_cluster-cell_type-cellchat-' + pathway]["communication_matrix"]
ccc_pvals = adata_disthr.uns['commot_cluster-cell_type-cellchat-' + pathway]["communication_pvalue"]

I found that even though prior data processing/normalization/analysis is uniform across samples, the CCC scores for the same (pathway, cell type 1, cell type 2) combo have quite distinct value distributions across different samples. CCC scores that are similarly significant (p-value = 0) can be very different in terms of magnitude across samples, making subsequent analysis and visualization difficult.

Therefore, I have the following questions:

Is it meaningful to compare the magnitude of CCC scores for the same cell type combination across samples?

• If not, what will be a good way to compare the COMMOT results across different samples?
• If so,
  • What is the possible range of communication_matrix?
  • Is communication_matrix normalized? Would you advise performing normalization/scaling to the scores before comparison? Should the scaling be on a sample level, on a (sample, pathway) level, or on a (sample, cell type 1, cell type 2) level?

Any advice will be much appreciated. Thanks again for your help!

[ysbioinfo]

I am facing the same problem. Wondering if there is any good solution

[ccruizm]

Same for me. Any thoughts on this would be incredibly helpful!