Hi,

I want to normalize gene expression counts data (from normal-contaminated tumor samples as well as normal samples) with VST (from DESeq2 package). Then I want to correct for tumor purity on these normalized data by fitting a linear model based on experimentally determined tumor purity estimates.

Do you have a suggestion on how to transform data back to raw counts after performing this correction (if it is possible at all)? Raw counts are needed to run a standard DESeq2 workflow on this. I could proceed with other packages with normalized data but I would like to use DESeq2 with raw counts. So my question is if it is possible to inverse the VST (e.g. by using some parameters) after performing the purity correction.

P.S. Concerning modeling tumor purity as a covariate just a remark in advance: In my current understanding, it is not justified to put tumor purity as an additional covariate in the design of DESeq2, as DESeq2 uses a log2 link-function and the covariates are modeled at this scale, while the tumor purity is already having an additive effect on the normal scale.

Thank you!

Best Regards

We don’t recommend modifying the counts in this way before DESeq2, as it breaks the link between the count and its precision.

I think if count scales with purity you could add in log transformed purity into the design to preserve that relationship.