I have RNA-seq data of whole adipose tissue (adipocytes were not purified) and other clinical measurements.
When I run WGCNA analysis, I noticed that the modules that correlate with the features of interest are exclusively found in different cell types (in this case, it seems a positively correlated module are genes from immune cells, while negatively correlated modules are from adipocytes).
In that sense, I feel like the only solid conclusion I can get is that the proportion of these cell types change with the clinical feature of interest. This is an issue, as I am more interested in adipocytes.
My questions are:
1- Do you agree with this conclusion?
2- How do other researchers deal with data of whole tissue RNA-seq?
3- Is purifying the samples and redoing the RNA-sequencing the only alternative or is there a statistical solution? I assume rerunning the analysis with the currently available data using only genes expressed in adipocytes is not justifiable.
The term you are looking for is (cell-type) deconvolution.
I am familiar with deconvolution. However -please correct me if I am wrong-, from what I understand deconvolution can give you cell type proportion but not what how gene expression changes on different conditions in a specific cell type unless you already have the signature of cell type A in condition X and the signature of cell type A in condition Y, which is what I am interested in. Cell type proportion is not interesting for me in this case.
Although I am not sure to which extent you can believe those, there are deconvolution algorithms that will give you not just the proportions, but also try to approximate the expression levels per cell type. Not my comfort zone though!
I will look into it. Thanks!