I am working with RNA seq data. I would like to compute the residuals with DESeq2 to remove confounding effect in the data before running WGCNA.

We computed the residuals by using  DESeq2 on our RNA Seq count matrix with this code:

fitted.common.scale = t(t(assays(dds)[["mu"]])/sizeFactors(dds))

counts(dds, normalized=TRUE) - fitted.common.scale

WGCNA suggests to log2 transform count data (from RNA-seq) before the analysis.

We noticed that the residuals are computed in the scale of the counts.

Is it correct to log2 transform the residuals (computed as described above) before using them in WGCNA? Or this would be a mistake?

Thank you in advance for your help.

For any method that can't handle count data directly, you should follow the recommendation in the DESeq2 manual and use either the rlog or variance stabilizing transformations to get log2-scale data that is approximately homoskedastic.

If you want to remove batch effects, first shoot your data through DESeq2's vst, then use that log2 like matrix with limma's removeBatchEffects function