Make the normalized counts from treatment relative to control-EdgeR

Question

[Pamela]

Hi. I run a DEGs analysis with EdgeR to compare two Groups.

Group 1 has control samples (n=3) and paired treatment samples (n=3). Same for Group 2. Therefore I have 12 samples in total.

Basically I run this:

sampleGroup=Group2_treatment
sampleGroupBaseline=Group2_control
refGroup=Group1_treatment
refGroupBaseline=Group1_control

As an output of EdegR I got 12 columns with the normalized counts for each sample.

My question is: What would be the best way to get the treatment normalized counts relative to the control normalized counts for each subgroup? Whether to use a ratio (/) or a difference (-) or something else?

I want to then use these lists to run GSEA, make heatmaps, etc.

Thanks for your help!

Pamela

Answer 1

[ATpoint]

Most commonly for heatmaps one uses the relative expression, here is why and code suggestions: Scaling RNA-Seq data before clustering? It is uncommon to get counts "relative to control". If anything you could show fold changes from the differential expression, but fold changes can be large due to noise, so showing them without stats can be problematic.

For GSEA, e.g. as in the camera method from limma you could use the logcpms from edgeR.

[Pamela]

Thanks for your answer. If it is not common to do that then how can I make a PCA or hierarchical clustering of the treatment samples "normalized" to the control? I will explain more in detail:

I see that the treatment samples cluster very nicely in two Groups, but I do not see that for the controls. I would like now to know whether the treatment "normalized" to control samples cluster in these same subgroups (or others).

How could I do that? Thanks

[ATpoint]

As I said, the concept of "normalized to control" does not really exist, or at least I have not widely come across it in my ~10 years of doing omics. PCA typically is done on normalized data on log2 scale, e.g. logCPM from edgeR. This will tell you how data cluster. You can color plots be treatment and control and see whether this tells you anything. I recommend to reconsider your idea of "normalized to controls", and read some omics papers first, getting a feel what sort of analysis and plots the field commonly does.

[Pamela]

Yes, I understand that this kind of analysis isn't typical, but I am trying to make sense of some data that is very heterogeneous. Thank you for your guidance, it helped me realize that the way I was expecting to go isn't the right one. Best wishes.