Hi,

I would like to perform deconvolution on bulk RNA-seq data, by using a reference matrix obtained from CELLxGENE. The dataset I want to use as a reference combines data from several studies, so there are multiple donnors, assay technologies, etc. I filtered my data by tissue, dissease and assay, and I end up with a subset which contains multiple donors from a few different studies.

The deconvolution tool I plan to use recommends the use of unnormalized and untransformed count data, so raw expression matrix.

My question here is: what is the right way to perform the batch correction? Should I do it before deconvolution, on expression counts, by using e.x. ComBat-seq (or would you recommend another tool for R?) ? Or shoud I instead control batch in the regression model applied to deconvolution results? This answer here led me to the latter option, but I am not sure what to do.

It may be trivial question but I lack experience, and I would greatly appreciate any advice and guidelines. If you need more information, like the dataset in question, etc., I will be happy to link it in the comments. Thanks!