Let's assume that I have two conditions and I would like to do a differential miRNA expression analysis between these two groups. I have log2 "reads per million" (RPM) values as my input data set. Now, I am wondering whether I should test the differential expression between these groups using a Mann-Whitney-U test on log2 RPM values or feed these reads to a edgeR/limma.voom/DESeq and run the regular RNASeq differential gene expression pipeline on them as if they are RSEM values?
I would suggest the limma/voom route unless you don't plan on sharing information across genes. I should also note that you should read through this thread on the Bioconductor email list regarding TCGA data and limma/edgeR/etc. There's also mention in that thread of the fact that the "normal" controls in TCGA are often far from being an appropriate comparison group.