[BioC] 2 issues about enriched gene sets via Roast

[Gordon K Smyth]

> Date: Thu, 23 Jan 2014 17:27:46 -0500
> From: <julie.leonard at syngenta.com>
> To: <bioconductor at r-project.org>
> Subject: [BioC] 2 issues about enriched gene sets via Roast

...

> On the other hand, when I ran roast for the contrast with few genes 
> differentially expressed, I got few gene sets enriched.  But what's odd 
> is it did find some gene sets enriched with FDR.Mixed < 0.05, but none 
> of the genes in the gene set were differentially expressed.  Are these 
> enriched gene sets false positives?  I'm not sure what's going on here.

This is in a way the whole point of gene set testing, to be able to pool 
information between genes to find an overal trend that is not apparent 
merely by looking at the individual genes.

The original GSEA publication (Mootha et al, Nature Genetics 2003) began 
by saying that no genes were individually significant in their data:

"When assessed with ... analytical techniques that take into account the 
multiple comparisons implicit in microarray analysis, no single gene had a 
significant difference in expression between the diagnostic categories 
(data not shown)."

They then went on to show that GSEA nevertheless found significant gene 
set changes.  So the phenomenon that you have observed was the original 
justification for the whole gene set testing approach.

Best wishes
Gordon

...

> Please advise.
>
> Thanks,
> Julie
>
>
> Julie Leonard
> Computational Biologist
> Global Bioinformatics
> Syngenta Biotechnology, Inc.

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