[BioC] DESeq: residuals instead of read counts

[Tuuli Lappalainen]

I've been using DESeq for RNASeq data analysis of differential 
expression of genes and exons. However, I have a dilemma: Instead of 
using read count data, I would like to use residuals of linear 
regression (after rescaling), but I'm unsure whether this is a valid 
thing to do.

The reason to use residuals is the correction for variation in insert 
size between lanes (as e.g. in Montgomery et al. Nature 2010) - or 
potentially some other confounders as well. I have been planning to 
rescale the residuals to have all the values >0, and the median across 
the samples per gene the same as in the original count data, and use 
these data then similarly to raw counts. Would this be a proper way to 
analyze the data?

Many thanks in advance,

Tuuli Lappalainen
Department of Genetic Medicine and Development
University of Geneva Medical School