RNAseq expression data log2 transformed has negative values.

Question

[juncheng]

Hi,

I got a fpkm normalized RNAseq gene expression data. Really a lot of genes at some groups have expression value less than 1, that means after log2 transform, the value is negative. Some of the values even be -1000, which is really annoying.

How do you usually treat with this values, from my experience, log transformed RNAseq expression data never have negative values.

[Kevin Blighe]

Update October 15, 2018

Just to clarify something for others arriving here: logging RPKM or FPKM values does not make these any better for conducting statistical comparisons. With no cross-sample normalisation used when producing RPKM / FPKM, these units are not suitable for differential expression.

Please read this: A comprehensive evaluation of normalization methods for Illumina high-throughput RNA sequencing data analysis

The Total Count and RPKM [FPKM] normalization methods, both of which are still widely in use, are ineffective and should be definitively abandoned in the context of differential analysis.

Also, by Harold Pimental: What the FPKM? A review of RNA-Seq expression units

The first thing one should remember is that without between sample normalization (a topic for a later post), NONE of these units are comparable across experiments. This is a result of RNA-Seq being a relative measurement, not an absolute one.

Answer 1

[dariober]

In these cases it's not unusual to add a pseudocount of 1 to all counts so genes with 0 return to 0 after log. Obviously you have to assume that adding 1 doesn't bias much the initial non-zero counts.

(EDIT: Apologies, this was meant to be a comment to question of how to treat genes with 0 counts)

[Devon Ryan]

Just for reference, a value of 0.25 (by default, it's the prior.count option) is added by edgeR before calculating log2 rpkm or cpm values. Whether this is needed or not depends on ones goals.

[juncheng]

Thanks, this seams to be a log2(x+1) transform is common. +0.25 seams also fine depends on the data.

[Kevin Blighe]

While it may be common to log(x+1) for RPKM / FPKM, this does not mean that it is statistically valid if your goal is differential expression analysis.

Answer 2

[Devon Ryan]

You treat negative values the same as positive ones. There's no reason to expect FPKM or TPM or CPM or even normalized counts to be greater than 1.

[juncheng]

Thanks. How about the 0 values, after log transform, they become inf, this should just be treated as NAs?

[Devon Ryan]

That's what I'd do.

[juncheng]

Thanks. I will do that