For example, for one gene:
logFC edgeR = 4 log2(mean reads condition A / mean reads condtition B ) = 3
For all genes, all the logFC computed from edgeR seems to be different ~ +1 from naive computation based on reads log2(mean Reads samples cond A / means Reads samples cond B).
It seems a huge difference, quite difficult to explain to a lambda user. Also curious to understand what could be the factor responsible for this kind of offset.
How can you explain that with edge R and glmfit ?
I'm doing something classical :
cm <- makeContrasts(contrasts = comp,levels = dge$samples$group) dge <- estimateDisp(dge, de.design,robust=T) fit.y <- glmFit(dge, de.design) lrt <- glmLRT(fit.y,contrast = cm)