edgeR: without replicates

[mrfox]

Hi all,
I have two RNA-Seq samples (normal and treatment) but neither has replicates. In such a case one option is to apply edgeR with Poisson model on our data. But I know that the latest edgeR can do a better job and somehow estimate the biological variation. Unfortunately, I do not know exactly how to implement this. Could anybody help me for this?

Similar questions have been discussed here:

http://seqanswers.com/forums/showthread.php?t=4055

[Gordon Smyth]

Estimating biological variation without replicates is tough. edgeR offers the following solution however that will work well when the proportion of transcripts that are differentially expressed is no more than 30%. The solution is part of the new code to fit generalized linear models and to handle multi-factor experiments.

Suppose that y is your DGEList object. Setup a null design matrix with only an intercept term, then estimate the dispersion robustly from your two samples:

design0 <- matrix(1,2,1)
y <- estimateGLMCommonDisp(y,design0,method="deviance",robust=TRUE)

Then you can use the resulting dispersion estimate y$common.dispersion for your subsequent differential expression analysis.

We answer questions more regularly on the Bionconductor mailing list than on SEQanswers, so you could post questions there for more details.

Best wishes
Gordon