Question

Differential profile for only one normal - tumor pair

1

Entering edit mode

takumima ▴ 10

@takumima-16365

Last seen 10 months ago

Italy

Hi all,

I analysed the list of DEGs in 10 tumour-normal pairs (without replicates), with edgeR. I used the paired design

design = model.matrix(~Patient+Tissue)

Now, I would like to get the list of differentially expressed genes for each tumour-normal pair, individually. The aim is to get the patient-specific DEG list for the drug discovery in the precision medicine.

edger degs paired design • 1.3k views

ADD COMMENT • link updated 6.5 years ago by Gordon Smyth 52k • written 6.5 years ago by takumima ▴ 10

score 3 · Answer 1 · 2018-07-05

3

Entering edit mode

Gordon Smyth 52k

@gordon-smyth

Last seen 8 hours ago

WEHI, Melbourne, Australia

If you've already analysed the paired design, then you will have a DGEList object with dispersion estimates stored in it. I will call this object dge.

Step 1: Make a new design with patient-specific effects:

design2 <- model.matrix(~Patient + Patient:Tissue)

Step 2: Fit the new glm using the trended dispersion previously estimated with the original design matrix:

fit <- glmFit(dge, design2, dispersion=dge$trended.dispersion)

Step 3. Now you can extract patient-specific effects. To get DE genes for the first Patient:

lrt <- glmLRT(fit, coef=11)
topTags(lrt)

Note: do not attempt to re-estimate the dispersion using design2. The patient-specific design matrix does not have any replication.

ADD COMMENT • link 6.5 years ago Gordon Smyth 52k

0

Entering edit mode

Thank you, Gordon, for your suggestions. Your solution seems to work!

However, I'm not sure if my code returns the list of DEGs in the tumour sample. This because by performing the analysis on a publicly available dataset, my gene of interest is up-regulated, exactly the opposite of what is reported in the paper.

gr = as.factor(c("N","T"))
groups <- rep(gr, times=10)

Patient <- factor(c("pat1","pat1","pat2","pat2","pat3","pat3","pat4","pat4","pat5","pat5","pat6","pat6","pat7","pat7","pat8","pat8","pat9","pat9","pat10","pat10"))
Tissue <- factor(groups)

design = model.matrix(~Patient+Tissue) #compare the tissues (tumor-normal) adjusting for any baseline differences between the patients by fitting an additive model

genefit <- glmQLFit(geneList, design)
geneqlf <- glmQLFTest(genefit)

geneqlf

#$comparison
#[1] "TissueT"