[BioC] Identifying genes that contribute to limma's roast/romer gene set enrichments

[Gordon K Smyth]

Dear Steve,

The genes that contribute most to the roast result are the same genes that 
are ranked at the top in a standard genewise DE analysis, i.e., ranked by 
topTable().  Roast doesn't output the leading-edge genes because they are 
already available from the standard genewise analysis.

The z-score is computed from the moderated t-statistics using zscoreT(). 
The roast help page says that the cutoff for PropDown and PropUp is z < 
-sqrt(2) and z > sqrt(2).  The probability of this is:

   > 2*pnorm(-sqrt(2))
   [1] 0.1572992

So the genes that contribute to PropUp and PropDown are those that have 
p-value less than 0.1572992 a standard genewise DE analysis.

There is no exact equivalent for romer, but obviously looking at the 
genewise DE results will still reveal the most important genes.

Best wishes
Gordon

> Stephen Hoang stephen.a.hoang at gmail.com
> Fri Jan 31 21:58:26 CET 2014
>
> Hi All,
>
> Is there a way to identify the genes that contribute to the gene set 
> enrichments calculated by roast/romer? I realize that all of the genes 
> in a given set are contributing to the p-value calculation, but is there 
> a way to get at, for example, the genes that contribute to the "PropUp" 
> value in the output of roast? And is the analogous gene set knowable in 
> the case of romer? I see in roast help page that the "up" and "down" 
> genes are categorized as such based on a z-score threshold, but what are 
> the values in this distribution -- global fold changes, fold changes 
> within a gene set, or something else altogether?
>
> Thanks in advance,
>
> Steve

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