[BioC] multiple comparison adjustment of p values in LIMMA

[Shi, Tao]

Dear Gordon,

Thank you very much for your quick reply!  Somehow I have missed the glaring statements on this issue in the decideTests help file.  My bad!

In both the help file and the user guide, you recommend to use method="global" when there are a few related contrasts.  However, from your 2nd statement below: "adjustments within a gene might not be needed 
when controlling FDR", are you suggesting to just use method="separate" instead?  I'm a bit confused.  I tried hard searching for your original post on this in BioC list, but 
didn't find it.

Thanks!

Tao
  



----- Original Message -----
From: Gordon K Smyth <smyth at wehi.EDU.AU>
To: "Shi, Tao" <shidaxia at yahoo.com>
Cc: "bioconductor at r-project.org" <bioconductor at r-project.org>
Sent: Tuesday, September 17, 2013 5:16 PM
Subject: Re: multiple comparison adjustment of p values in LIMMA 

Dear Tao,

Would you please consider reading the documentation for decideTests?  If you did that, you would know that limma does in fact offer options for doing multiple comparison adjustments across contrasts within a gene as well across genes.

I have also previously explained on this list why adjustments within a gene might not be needed when controlling FDR.

Best wishes
Gordon

On Tue, 17 Sep 2013, Shi, Tao wrote:

> Dear Gordon and list,

> Could you please explain why the multiple comparison adjustment procedure implemented in limma (i.e. decideTests) only do adjustment across genes but not across treatment contrasts within a gene?   I found one of your earlier replies related to this here ( https://stat.ethz.ch/pipermail/bioconductor/2012-November/049385.html ):

> "Post hoc tests are done in limma using decideTests(), and many options are offered.  You won't find classical methods like TukeyHSD though, because limma isn't doing classical Anova and because methods like TukeyHSD don't generalize well to high-dimensional datasets like microarrays."

> Could you please elaborate on this?  Take a 1000-gene data set with 3 treatment groups as an example, if you're doing all 3 pair-wise comparisons, it's 3000 tests.  One would think that controlling FDR for all 3000 tests is quite different from controlling FDR for 1000 tests per comparison.  If I only have a dataset with 50 genes, does this change your statement I cited earlier?

> Thank you very much!
> Tao
> 

______________________________________________________________________
The information in this email is confidential and intended solely for the addressee.
You must not disclose, forward, print or use it without the permission of the sender.
______________________________________________________________________