Hi all,
I was just wondering whether it would make sense to combine the upperquartile normalization with limma-voom analysis of RNAseq data.
this is just out of interest, not driven by the data or so. At first, I did
uq <- DGEList(counts=initDGE, group=responseStatus)
uq <- calcNormFactors(uq, method="upperquartile")
uq <- estimateDisp(uq, design)
uq.v <- voom(uq, design, normalize.method="none")
cat("Limma::voom & eBayes after UpperQuartile\n")
fit <- lmFit(uq.v, design)
v.UQ <- contrasts.fit(fit, contrasts=contrastMat.R)
v.UQ <- eBayes(v.UQ)
then extracting a toptable from this object.
Ofcourse I get a toptable, but is this a correct approach. first, I am generating upperquartile normalized counts, then I feed this to 'voom' to log-transform the data without further normalization by voom, then I do the modelfitting etc.
Thank you for your advice.
Thank you for taking the time to reply!
It was just my uncertainty of using the commands.
I am trying to compare different normalization methods with different statistical testings (basically TMM, upperquartile, quantile and DESeq2's default method across with RLT, QLF (both edgeR) and with voom-eBayes).
Comparing methods using experimental data is usually not very instructive. If you don't already know which genes are differentially expressed, and you get different results for each normalization method, what criteria are you planning to use to compare the different normalization methods?