DESeq2 vignette, the following snippet provides the example of how
cqn offsets can be used in the analysis:
cqnOffset <- cqnObject$glm.offset cqnNormFactors <- exp(cqnOffset)
Meanwhile, the EDAseq normfactors should be used after flipping their sign (note the
EDASeqNormFactors <- exp(-1 * EDASeqOffset)
The same usage is echoed in the documentation of these tools.
However, from my understanding both
EDASeq return offsets that are of the following form:
offset = log(normed_counts) - log(actual_counts)
See page 4 of cqn vignette (starting "the normalized expression values are", rearrange to move
y (equiv. to log2(actual_counts)) left of equation to get the expression above), and see section 7.3 of
EDASeq vignette which uses this expression as is.
I may be missing something very obvious here, but, given this, why multiply the
EDASeq offset by -1? Or, equivalently, why not do that for