Dear all,

I have a very basic question about PCA for RNAseq data. I am currently analysing data normalized with the voom method as described in the Limma user guide. Now I would like to explore the distribution of my samples using PCA with the prcomp function. However, I am not sure of the good way to use the scaling (scale.) function with prcomp. If I understand well, scale. = TRUE is used when we expect that a small number of variables with a high variance will mainly contribute to the total variance of the distribution, which is seen on raw read counts of RNAseq data. But the voom normalization should have reduce this effect. So, intuitively, I could use scale. = FALSE for normalized RNAseq data. Am I correct? If not, could anyone explain me why?

Please, do not hesitate to tell me if my explanations were not clear.

Thank you very much in advance.

Best,

Nico

If you want to explore your data, you could just use plotMDS on the EList object produced by voom. This generates a MDS plot that achieves the same thing as a PCA plot, i.e., libraries that cluster together are more similar, those that are far apart are more different.