Hi All,

I have a scRNA-Seq dataset for which I would like to regress out the effects of cell cycle and then perform a differential expression analysis. I believe cell cycle genes are dominating and masking interesting gene expression patterns which would otherwise be present in the diff expression output.

Seurat can be used to regress out cell cycle genes via their ScaleData() function. However, the generally accepted method to perform differential expression is to use raw counts (certainly for DESeq2) and I will therefore lose any effect of my cell cycle regression.

So my question is - can I somehow use my scaled data to perform differential expression after cell cycle regression?

Thanks.

In general DESeq2 takes any kind of variables to be adjusted for as covariates for its design, so like

`~cluster + cellcycle`

. Cellcycle could here e.g. be the cell cycle phase per cell. Alternatively, you could use the residuals of the regression with something like a t-test or Wilcox test. It depends, as usual. How did you identify the cell cycle bias? How did it manifest, so do you have clusters of basically the same cell type separated by cell cycle? I found it useful to simply exclude the genes that drive the separation into the cell cycle phases from the set of genes used for clustering, that solved all my problems related to cell cycle confounding while having no need for any kind of regression. Please elaborate, give details.