[BioC] Questions about edgeR

[Mark Robinson]

Hi Rabe,

Some comments below.


On 19.12.2011, at 10:57, Asma rabe wrote:

> Hi All,
> 
> I have RNA-seq data for different time points 0hr,2hr,4hr,6hrs and 8hrs.
> 
> I would like to use edgeR to get genes that are differentially expressed
> throughout the experiment.
> In a recent edgeR user's manual ,it has been suggested to use a dispersion
> value of 0.1 in case of no replicates for organisms that are genetically
> similar to humans.
> so it is possible to be used for mouse data for example,what about other
> organisms like yeast ??

Yes, it is possible.  But, as mentioned in the manual, none of these solutions are ideal.


> As exact test is used to get DE genes across 2 samples and i need to get DE
> genes across my time course experiment,i used GLM.
> 
> when i  used GLM model to find DE genes across my samples i ran the
> following commands:
> 
> mm<-model.matrix(~0+my_groups_factor)
> fit<-glmfit(my_data,mm,dispresion_value)
> fit<-glmLRT(data,fit)
> topTags(fit)
> 
> -----
> i got:
> 
> coefficient: group8
>                 logConc         logFC          PValue            FDR
> gene1         .....                 .....               ....
>    ....
> gene2         ..                       ....
> ....                   ....
> ..
> .
> ..
> 
> 
> ---
> I have a basic question(sorry i'm not statistician)
> Why the last gorup 8hrs was chooosen to be the coef.??

Have a look at the documentation for glmLRT and specifically the 'coef' argument:

?glmLRT

----
coef:     … By default, the
          last column of the design matrix is dropped to form the
          design matrix for the null model.
----

You have 5 levels for your time course (therefore, design matrix has 5 columns).  To test for any difference between groups, I would suggest reparameterizing your design matrix and then specifying the 'coef' argument.  Something like the following:

mm <- model.matrix(~my_groups_factor)
fit <- glmFit(my_data,mm,dispresion_value)
fit <- glmLRT(data,fit,coef=2:5)
topTags(fit)

Hope that helps.

Regards,
Mark


> 
> Thank you in advance.
> Best Regards,
> Rabe


----------
Prof. Dr. Mark Robinson
Bioinformatics
Institute of Molecular Life Sciences
University of Zurich
Winterthurerstrasse 190
8057 Zurich
Switzerland

v: +41 44 635 4848
f: +41 44 635 6898
e: mark.robinson at imls.uzh.ch
o: Y32-J-34
w: http://tiny.cc/mrobin