Relative log expression normalization without negative values

Question

[demoraesdiogo2017]

Hello. I would like to analyze a public RNA-seq data set which the author claims to be normalized with relative log expression normalization. However, in my experience, this generates negative expression values, but all values are positive in his matrix. Am I mistaken or is there a upstream processing not mentioned by the author?

Answer 1

[jared.andrews07]

RLE does not return negative values, it divides the raw counts by the size or normalization factors.

I expect you are thinking of log2 normalization, which will return negative values for counts below 1. This is why it's usually performed as log2(normalized counts + 1), e.g. the normTransform function in DESeq2.

[ATpoint]

I think OP assumes that logcounts were standardizes, e.g. Z-scored so that would be relative expression, but I agree that here (based on how the counts look in the file of this GEO entry) that it is probably RLE-normalized values (that is the DESeq2 normalization method @OP).