[BioC] Using EdgeR to test for null ratios other than 1:1

Gordon K Smyth smyth at wehi.EDU.AU
Thu Jan 2 23:28:54 CET 2014

Dear Emily,

Well, you can test null ratios other than 1:1 using contrasts.  For 
example, if Group is a factor with two levels, then

   design <- model.matrix(~0+Group)
   fit <- glmFit(y, design)
   lrt <- glmLRT(fit, contrast=c(1,-1.5))

would test against the 1:5:1 null.  To test 1:5:1 for one particular 
gene, you might use:

   lrt1 <- glmLRT(fit[gene1,], contrast=c(1,-1.5))

To test 3:4 for another gene, then you might use:

   lrt2 <- glmLRT(fit[gene2,], contrast=c(4,-3))

As you can see, it is quite straightforward.

However if there are many genes for which the null ratio is other than 
1:1, then it would be better to build the null ratio into the offsets 
instead.  This also has implications for normalization so, if this is 
relevant to you, then it would be better for you to explain more fully 
your intended application.

Best wishes

> Date: Thu,  2 Jan 2014 01:32:36 -0800 (PST)
> From: "Emily [guest]" <guest at bioconductor.org>
> To: bioconductor at r-project.org, em6 at ebi.ac.uk
> Subject: [BioC] Using EdgeR to test for null ratios other than 1:1
> I was wondering whether I could use EdgeR to test for differences 
> between two groups with an inherent bias in their reads proportional to 
> each other. By this I mean that instead of each gene being tested with 
> the null of 1:1 could I modify edgeR to test for a gene where the null 
> is say 1.5:1 and another gene with a null of 3:4 etc.
> Many thanks in advance,
> Emily
> -- output of sessionInfo():
> ---
> --
> Sent via the guest posting facility at bioconductor.org.

The information in this email is confidential and intend...{{dropped:4}}

More information about the Bioconductor mailing list