Change of cell proportions or differential expression?

Question

[Aspire]

Suppose bulk RNA-Seq of two conditions is given, together with the gene-expression profile typical of various cell types.

Cell type proportions can be deconvoluted from bulk data (more easily so with a reference) eg via CibersortX.

A person wants to know whether the changes she observes between the two conditions are due to a change of cell type proportions between the two, or differential expression.

• Is it possible to distinguish between a change in cell type proportions (the cell types composition is different between the two conditions) and a change in gene expression (both conditions have the roughly same cell types proportions, but the gene expression pattern has changed between the conditions )

• How is it possible, conceptually?

[Aspire]

Posted originally on stackexchange, was advised that a bioinformatics centered forum is more appropriate

[rfran010]

I don't think it is possible. You can make assumptions though.

However, these are also related concepts where if you have Population A with 10% progenitor and 90% neutrophils and the Population B with 30% progenitor and 70% neutrophils, this is a difference in cell proportion due to differentially expressed genes (progenitor programs are upregulated in progenitors and differentiation programs are downregulated in neutrophils).

So it really just depends on if cellular identity is perturbed, where DEGs are disconnected from cell identity, or if specific programs are being upregulated, e.g. apoptosis, then they would be due to "a change in gene expression". This would require assumptions to distinguish. Maybe more informative would be to run gene ontology or enrichment.

[Aspire]

It seems that CibersortX claims to be able to distinguish between a change in cell-type proportion, and between change in gene expression between two conditions. If one has a signature matrix, one can input that signature matrix as well as the bulk RNA-Seq to the tool, and it is able to yield

(1) The proportions of each cell-type, per-sample.

(2) The expression profile per cell-type (per condition).

From the latter, differential expression per cell-type can be performed.

Comparing those two results two can enable to answer the question imho, eg if the cell-type proportions between the two conditions have not changed, then the story is about DE.

Not that I have a clue what is the "magic" behind the calculations enabling this...