DEGseq or EdgeR

[MerFer]

Hi,
I'm using EdgeR to analyse RNA-seq data with replicatons. But now I want to compare two conditions with no replications. But I am not sure, which one (DEGseq or EdgeR) better to use without replicatons.

Thanks

[Simon Anders]

Hi

please have a look at this thread where the question was recently discussed:

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

In addition to what I've written there: DEGSeq assumes the variance to be at the theoretical minimum as given by the Poisson distribution. This has turned out to be not a tenable assumption and leads to way too optimistic p values, especially with higher counts. EdgeR, as well as our tool, DESeq, estimate the variance from the data, and both tools now offer a mode to do this even when no replicates are present (see manuals for caveats).

Simon

http://www-huber.embl.de/users/anders/DESeq/

[kryskelly]

There is also the baySeq package for differential expression analysis.
Krys

[Simon Anders]

[My answer that I typed yesterday somehow got lost, so I try again:]

Have a look at this thread where the same issue was discussed.

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

A remark in addition to what I wrote there: You will find that the number of hits called by DEGSeq and edgeR differ drastically. This is because edgeR estimates the noise from the data while DEGSeq assumes implicitly that it is at the theoretical minimum as given by the Poisson distribution, an assumption that is not realistic (see Nagalakshmi et al., Science, 2008).

As another alternative to edgeR, I would, of course, like to advertise our own Bioconductor package, DESeq, which estimates the noise from the data as well but in a different way. See http://www-huber.embl.de/users/anders/DESeq/

Simon