Hello,

I am doing a pilot study using bulk RNAseq data for differential expression in human brain tissue from healthy controls and Alzheimer's patients. I performed a cell type deconvolution analysis to determine if the control samples and Alzheimer's samples have a different proportion of neuronal vs non-neuronal cells. The Alzheimer's tissue shows a consistent trend of having ~10% less neuronal cells.

I know it would be possible to add a continuous covariate such as percent neuronal cells to the DESeq2 differential expression analysis, but I was wondering if this approach seems sound? Would it be better to add a covariate for percent neuronal cells or try to do some kind of normalization using genes only expressed in neurons?

This is a pilot to a much larger study we are planning on doing and I am trying to make sure the differential expression we find is not due to neuronal death associated with Alzheimer's.

Any input will be much appreciated!

```

I have used composition as a control factor before. It's better to add covariates than to try to manipulate the counts.

There are methods like BMIND that attempt to do DE on the CTS expression (inferred). Worth looking into.