Hi so I notice something interesting.
data <- voom(x, plot=T, design = design2)
vfit <- lmFit(data)
vfit <- eBayes(vfit)
topTable(vfit,coef=2,sort.by="P")
vfit <- lmFit(data$E, design = design2)
vfit <- eBayes(vfit)
topTable(vfit,coef=2,sort.by="P")
The two results above gives slightly different p value - very slightly but enough that the top10 from topTable gets shuffle a bit. Does anyone know why that is the case? thanks!
The difference between your two examples are that you are using the complete Elist object from the voom function in the former and only the E slot of the same list in the later. By using only the normalized expression values you are not using all information from the voom function (see https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4053721 ). The weights can be used in the linear modelling, but it will also work without them. Without the weights the results can of course be different.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
@thokall oh yes that right. I read it more carefully now whereby I thought previously that the E value were not only normalized but weighted as well, my bad, thanks!