Dispersion & FDR in edgeR
1
0
Entering edit mode
Lisa • 0
@lisa-17987
Last seen 5.4 years ago

I am currently using the edgeR package for RNASeq analysis, however, I have a bit of a problem. In the current dataset that I am working on, I noticed that following calculation of the tagwise dispersion that the FDR for all of the genes is 1. However, this doesn't occur after estimating the common and trended dispersion. Any feedback on as to why this may be occurring is greatly appreciated. I am using edgeR version 3.22.5. Many thanks, Lisa. 

 

 

edger R • 688 views
ADD COMMENT
1
Entering edit mode
Aaron Lun ★ 28k
@alun
Last seen 26 minutes ago
The city by the bay

When you get all FDRs equal to 1, the interpretation is pretty clear - you cannot detect any DE genes at any FDR threshold. This is not necessarily wrong, maybe there's just no DE genes between your conditions.

Using the common and trended dispersions will tend to be anticonservative as the true dispersions for highly variable genes are underestimated. This understates the likelihood of stochastic differences between conditions, increasing the number of (spuriously) small p-values and false positives. The tagwise dispersion is closer to the true dispersion and gives a more accurate reflection of the significance of DE.

(Conversely, the dispersions for lowly variable genes will be overstated when you use the common and trended estimates. This manifests as increased conservativeness for these genes but not in a manner that cancels out the anticonservativeness from the effect described above.)

You don't show any code, but it sounds like you're using the LRT methods. I would suggest switching to the quasi-likelihood methods (glmQLFit and glmQLFTest), which are more accurate in terms of type I error control.

ADD COMMENT
0
Entering edit mode

Thank you so much for your feedback. And yes you are correct, I am using LRT methods. I hadn't given the quasi-likelihood method a try but I will now do so. 

ADD REPLY

Login before adding your answer.

Traffic: 771 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