edgeR - normalized counts from multiple datasets

Question

[druggable]

Hi,

I am combining multiple independent rnaseq datasets from the same species but different conditions and analyzing them together. I normalized the counts per dataset using calcnormfactors in edger. I will have to do a second normalization when the different datasets are combined. I rounded up the normalized counts. Is it valid to use these rounded normalized counts to another round of calcnormfactors?

[1561775911]

Hi, Kevin, I am studying a mRNA predictor for immune-checkpoint inhibitors response across varible types of tumors.When I get different kinds(population with diffrent tumors) of RNA-seq count matrices) ,I determined to choose some HouseKeeping genes to normalize the data.What do you think of this method to preprocess the data?

[Kevin Blighe]

Oh, why do you not normalise as per the recommended methods for RNA-seq? Housekeepers are used for IHC, PCR, etc. There is no true 'housekeeper' whose expression remains stable, and I imagine that it's even less stable in tumour cells.

[1561775911]

As I know, the methods of normalization for RNA-seq count include quantile normalization of limma-voom or edgeR.I can't confirm that these methods are suitable for RNA-seq count which will be used for the subsequent machine learning to find the predictors?

[Kevin Blighe]

For downstream machine learning, I think that logCPM counts from EdgeR or r-log counts from DESeq2 would be fine. If you can obtain log2 counts from Limma/Voom, then good too.

[1561775911]

I have done the TMM and logCPM with code as follow:

library(edgeR)
dge<-DGEList(counts = as.matrix(exprs))
dge_TMM<-calcNormFactors(dge)#TMM
logCPM <- cpm(dge_TMM, log=TRUE, prior.count=3)#log2 transformation

But it confused me that when I compare the exprs matrix with the dge_TMM$counts,they are the same!Does the TMM process not change the exprs matrix??

[Kevin Blighe]

You mean that dge_TMM$counts is the same before and after you run cpm()?

[1561775911]

Hi,Kevin, I mean dge_TMM$counts is the same as the exprs which confused me.As I know,calcNormFactors(dge) is the process of TMM,but the dge_TMM$counts is not changed compared with the exprs

[GenoMax]

Please use ADD COMMENT/ADD REPLY when responding to existing posts to keep threads logically organized. This comment belongs under @Kevin's answer.

[1561775911]

Hi, genomax, I want use the 'ADD REPLY', but when I click on it ,there is no response and I can only submit answer .What's more, 'ADD REPLY' button on my page is gray(I don't know which color is on your page ,but it seemed that it is invalid on my page).Could you please tell me the reason? #ps. browser that I use is firefox.

[GenoMax]

We have seen this symptom in past for people using biostars from some parts of Asia. Not sure if it applies to you. ADD REPLY/ADD COMMENT buttons are indeed gray in color and are clickable for rest of us.

[1561775911]

It works when I use the browser of chrome.You can share the experience of mine to others who does not use the browser of chrome

[GenoMax]

Thanks for letting us know.

Answer 1

[Kevin Blighe]

You should only do one round of normalisation with all of your samples combined, and in your design formula you should include 'batch' or 'study' as a covariate. My recommendation for you is to actually use DESeq2, as I know for definite that it deals with these issues of batch and multiple experiments quite well.

Kevin

[druggable]

Thank you very much for your answer and recommendation.

[Kevin Blighe]

Okay. Keep in mind that, by including 'batch' in the design model, you are only accounting for batch in the statistics. This will not modify your normalised counts.