Hi,
I understand that voom normalizes the raw count data to log2 cpm data and then estimates the mean-variance relationship from the normalized data.
I want to know if there is strong bonding between log2 cpm data and estimation of mean-variance relationship? In other words, will I get sensible results if I modify the voom code such that it accepts normalized data such as TMM normalized data and then it estimates the mean-variance relationship from TMM normalized data?
I am developing a pipeline and want to implement voom. But, I want to give users option to input normalized data (with their choise of normalization method) if they wish. However, I am not sure if mean-variance relationship will still be as good as that calculated from log2 cpm data.
I will appreciate for any information in this regard!
Thanks in advance!!
To add to James' answer; the voom authors probably didn't want to support arbitrary normalization schemes, because there's no guarantee that the users will normalize in a sensible manner that preserves the mean-variance relationship. See A: Differential expression of RNA-seq data using limma and voom() for a discussion of what happens when, say, FPKMs are used for normalization. Sometimes, having more flexible code just gives your users more rope.