Comapre two groups in edgeR- Coefficient and logFC explanation
1
0
Entering edit mode
JG13 • 0
@eeb93914
Last seen 4 months ago
Greece

i have a dataset in which i want to compare the disease vs control (disease/control) When I am doing the analysis I have Coefficient: -1*control 1*disease. Is this correct to compare disease vs control ? and the total DE results are

 summary(deg1)
       -1*control 1*disease
Down                     5342
NotSig                  58145
Up                       5462 

Could you please explain what is going on with the logFC?

Coefficient:  -1*control 1*parkinson 
                     logFC   logCPM        LR       PValue          FDR
ENST00000309758.6 5.596663 5.290097 104.58809 1.503581e-24 1.036704e-19
ENST00000375650.5 2.499489 9.443004  73.07818 1.246143e-17 4.296015e-13
ENST00000375651.7 2.814114 9.399579  71.41025 2.901603e-17 6.668753e-13
ENST00000525876.1 3.499724 2.089409  70.70220 4.154286e-17 7.160846e-13
ENST00000674129.1 -1.615221 1.126     12.388    0.00043209  0.00730928

if it is positive means overexpression in disease or not? and if it negative means underexpression ?

edgeR • 792 views
ADD COMMENT
2
Entering edit mode
Yunshun Chen ▴ 870
@yunshun-chen-5451
Last seen 21 days ago
Australia

It looks correct if your design matrix is constructed in a way that the disease column represents samples in the disease group and the control column is for all the control samples.

Under this testing contrast, a positive logFC means up-regulation in parkinon (disease) compared to the control, and vice versa.

ADD COMMENT
0
Entering edit mode

So in case where i have Coefficient: -1*disease 1*control the positive logFC means underexpression and the negative over in disease? so it is the opposite than before ?

ADD REPLY
1
Entering edit mode

You can interpret this as if it were simple algebra (which it is). -1 * disease 1 * control is identical to control - disease, which shows the directionality. This is not how I would normally fit a contrast, because a negative logFC indicates upregulation in disease, which is not IMO how people would normally think about such things. I usually put the 'least affected' group in the denominator (these are logged coefficients, so log(disease) - log(control) == log(disease/control)), so control is always subtracted from treatment or disease or whatever. In which case the contrast should be 1 * disease -1 * control.

ADD REPLY
0
Entering edit mode

Yes i completely agree but somehow i used in all my data -1* disease 1* control which was coming from manual of edger and lrt12 <- glmLRT(fit, contrast=c(-1,1)). Thats why I am confused how to select the contrast in a proper way.

My design is :

$design
  disease control
1   0        1
2   0        1
3   0        1
4   1        0
5   1        0
6   1        0

So the -1 * disease 1 * control is the opposite .

the -1*control 1*disease or 1*disease -1*control is the same ?

ADD REPLY
0
Entering edit mode

Well, you shouldn't be using glmLRT these days. You want the quasi-likelihood pipeline.

I'll answer your question with a question. If contrast = c(-1,1) doesn't give you what you want (you want the opposite), how do you think one would change that to get what you do want?

ADD REPLY
0
Entering edit mode

I tried the contrast=c(1,-1)) which is the opposite. Maybe is should try the quasi-likelihood ! Thank you !

ADD REPLY

Login before adding your answer.

Traffic: 633 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6