Question

Can we use glmQLFit instead of glmFit and then glmLRT in edgeR

0

Entering edit mode

simarsidhu25 • 0

@simarsidhu25-12016

Last seen 7.7 years ago

I want to determine DGE analysis with my data that don't have any replicate. I just want to know that can I use glmQLFit and then glmLRT as I am not getting any DE genes by glmFit?

edger differential gene expression glmlrt() • 3.1k views

ADD COMMENT • link 7.9 years ago simarsidhu25 • 0

0

Entering edit mode

But, the exact test is only applicable to experiments with a single factor and I have more than one factor.

ADD REPLY • link 7.9 years ago simarsidhu25 • 0

0

Entering edit mode

Well, in that case, you can use glmFit() and glmLRT(), can you not?

ADD REPLY • link 7.9 years ago Gordon Smyth 51k

0

Entering edit mode

Yes, but I am getting single DE gene. But I am getting good results when I am using glmQLFit() and glmLRT(). Would it be possible?

ADD REPLY • link updated 7.9 years ago by Gordon Smyth 51k • written 7.9 years ago by simarsidhu25 • 0

0

Entering edit mode

No it is not possible. I don't know what you are doing (you don't give code) but it cannot be correct.

It isn't even possible to run glmQLFit() if you don't have replicates -- it will return an error.

ADD REPLY • link 7.9 years ago Gordon Smyth 51k

0

Entering edit mode

Thanks, I have got the point.

ADD REPLY • link 7.9 years ago simarsidhu25 • 0

score 4 · Answer 1 · 2016-12-13

You cannot use glmQLFit() if you don't have replicates.

In the absence of replicates, the only choices are exactTest() or glmFit() followed by glmLRT().

In the absence of replicates, you have to set the dispersion yourself. If you don't have any DE genes and you think you should, then apparently you have set the dispersion too high. This has nothing to do with trying to change the edgeR functions you are using.

score 0 · Answer 2 · 2016-12-13

glmQLFit/glmQLFTest can be used almost anywhere that glmFit/glmLRT can, but it's not going to give you a longer list of significant genes than glmFit/glmLRT. In fact, it's almost certain to do the opposite.

Performing a differential expression test with no replicates is a last resort method, and you should not be surprised when it gives you no significant results.