I'm using voom and limma to analyse RNAseq data. I'm comparing the percentage distribution of reads among different categories of transcripts within a sample with the aim of making a stacked percentage bar plot (or similar), for instance: counts for protein-coding (97%) vs pre-miRNA (1%) vs snoRNAs (0.1%) etc. I'm wondering whether I should be using the raw counts output from read summarisation (eg featureCounts output) or say the "E" matrix of normalised log2 counts from the EList object after voom transformation, or if it makes little difference? I'm unsure if the normalised counts can be used for purposes other than in the limma pipeline or if they should be multiplied by the normalisation factors first?
Any advice/experience much appreciated.