Using bulk RNA-seq deconvolution as covariates in bulk RNA-seq differential expression analysis

Question

[telroyjatter]

There are a few different methods out now that claim to estimate cell-type composition of bulk RNA-seq samples, including CibertsortX, DWLS, and others. Has anyone ever used the results from these methods as covariates when running differential expression analysis on bulk RNA-seq data? I suppose one question is whether differing cell type compositions between samples is an artifact during tissue collection (e.g., neurons, glia, astrocytes when punching brain tissue), or if it's a real result from experimental treatment.