Question

Using Limma R package For Proteomics differential expression

2

Entering edit mode

David Lyon ▴ 340

@david-lyon-4016

Last seen 3.1 years ago

United States

I spent many hours trying to find this answer, but alas to no avail. I have somalogic proteomics data normalized. I have 2 groups (test and control) I want to find the differential expression between those groups.

Normally for RNAseq I would do something like below.

Can someone show me exactly have to change the code used for RNAseq for proteomics.

Thank you very much in advance!

contr.matrix <- makeContrasts(
   res = test - control,
      levels = colnames(design))
contr.matrix

vfit <- lmFit(v, design)
vfit <- contrasts.fit(vfit, contrasts=contr.matrix)
efit <- eBayes(vfit)

as.data.frame(topTable(efit,coef = 1,number=Inf))

limma ProteomicsWorkflow • 9.3k views

ADD COMMENT • link updated 15 months ago by Gordon Smyth 52k • written 3.1 years ago by David Lyon ▴ 340

score 2 · Answer 1 · 2021-11-03

2

Entering edit mode

James W. MacDonald 67k

@james-w-macdonald-5106

Last seen 2 days ago

United States

At the most basic level what you have done is correct, and wouldn't change regardless of the underlying source of data. In other words, you don't fit a 'proteomics' model or an 'RNA-Seq' model. However, there are underlying assumptions for the model you are fitting, and you should make sure that they are being met, using things like plotSA and plotMD. If those look OK, then it's OK. If not you likely need to make changes, and what those changes should be is almost certainly beyond the scope of this support site. In that case you would do well to find a local statistician to help.

ADD COMMENT • link 3.1 years ago James W. MacDonald 67k

0

Entering edit mode

Thank you when I compare doing a Wilcoxon test Wilcox.test() with the Limma code above. I get ~ dozen of proteins that was significant and had been previously reported running Wilcoxon.test while running Limma with the above code gives zero proteins p.adj < 0.05. I was hoping to see Limma pro be just as sensitive or even more sensitive than Wilcoxon test. Do you think filtering out non-variable proteins ? and what code would you recommend to run here to improve the power of running Limma.

I did check those plots plotSA and plotMD and they both appear to be 'OK'

Thank you in advance.

ADD REPLY • link 3.1 years ago David Lyon ▴ 340

2

Entering edit mode

Assuming that the underlying assumptions for limma are met (which you are checking with plotSA and plotMD and could also use plotMDS to look for outlier samples), the model you fit with limma is more powerful than a Wilcoxon. So if you are getting significant results with a Wilcoxon and not with limma, then I would imagine that your data don't meet the assumptions of one or the other. Or perhaps you are not adjusting the Wilcoxon p-values for multiplicity, which limma is doing. If that is the case then you are not making a fair comparison.

This site is mainly intended to help people with technical issues with the packages, rather than helping with analyses, so I am afraid I can't help much more.

ADD REPLY • link 3.1 years ago James W. MacDonald 67k

score 0 · Answer 2 · 2023-09-11

0

Entering edit mode

samsiljee • 0

@fb510204

Last seen 15 months ago

New Zealand

Another thing to consider is how missing values are handled. I'm not sure about Somalogic, but certainly LFQ DDA mass-spec gives a lot of missing values which need to be considered. I personally use the MSstats package which was specifically designed for proteomics rather than limma which was designed for RNA-seq and microarray experiments.