[BioC] DESeq nbinomGLMTest specific comparisons

[Ryan C. Thompson]

In general, you would subset the CountDataSet to only include the 
samples from the two conditions you care about. Do this after 
estimating dispersions and before fitNbinomGLMs.

But more generally, I agree that the real solution is "use DESeq2".

-Ryan

On Thu 09 Jan 2014 03:41:39 AM PST, James Floyd wrote:
> Hi,
>
> Thanks for your reply, and sorry for the delay. As always, I got
> distracted by other projects Š
>
> I am looking to perform a DE analysis for a disease with three states,
> while correcting for smoking status. So I was using Š :
>
> fit0 = fitNbinomGLMs( cdsFull, count ~ smoking )
> fit1 = fitNbinomGLMs( cdsFull, count ~ smoking + condition )
>
> Š then comparing the models to see whether adding condition improves the
> fit of the model and has a significant effect (as described in the
> vignette).
>
>
> However, if I wanted to make a specific comparison of, say, the two most
> extreme disease states (state A against C, ignoring B), how would I go
> about doing this?
>
> Thanks for letting me know about DESeq2, I was unaware of it.
>
> Cheers,
> Jamie
>
>
>
> On 27/11/2013 17:21, "Steve Lianoglou" <lianoglou.steve at gene.com> wrote:
>
>> Hi,
>>
>> On Tue, Nov 26, 2013 at 8:21 AM, James Floyd <j.a.floyd at qmul.ac.uk> wrote:
>>> Hi,
>>>
>>> I notice in the DESeq
>>> vignette<http://bioconductor.org/packages/release/bioc/vignettes/DESeq/in
>>> st/doc/DESeq.pdf> there is a "Fixme" comment about including how to test
>>> specific comparisons using nbinomGLMTest (page 12). It would be great if
>>> someone with experience doing this could let me know how it is done.
>>
>> Can you be more specific of what you'd like to test? Perhaps you can
>> setup your problem for us and point out what particular comparison
>> you'd like to make, and someone can help you better.
>>
>> Furthermore, if you are using DESeq, you might consider shifting over
>> to DESeq2. It offers a number of benefits over DESeq, and (I believe)
>> this is where the DESeq authors are focussing the lion's share of
>> their attention anyhow.
>>
>> -steve
>>
>> --
>> Steve Lianoglou
>> Computational Biologist
>> Genentech
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor