[BioC] B statistic in limmaGUI

[Elizabeth Brooke-Powell]

James,

Thank you for that extra information, very well explained, and makes a lot
of sense.

Liz
-----Original Message-----
From: James Wettenhall [mailto:wettenhall at wehi.edu.au] 
Sent: 20 August 2004 08:05
To: Elizabeth Brooke-Powell
Cc: 'Sean Davis'; bioconductor at stat.math.ethz.ch
Subject: RE: [BioC] B statistic in limmaGUI

On Thu, 19 Aug 2004, Elizabeth Brooke-Powell wrote:
> We are
> interested in things that do not move. From our current
> understanding

Hi Liz,
The tests for differential expression in limma, limmaGUI etc.
are generally set up to test for differential expression, rather
than lack of it.  If you ask the question "How many genes are
not differentially expressed in one comparison?", the answer
would usually be thousands of genes!  By looking at the most
negative B statistics, you are not only looking at genes whose
expression levels are unchanged, but you could also be looking
at genes whose changes in expression are extremely variable
between replicate arrays (so they are not ranked as high as
genes whose changes in expression are perfectly consistent
between replicates).

It is natural to get noise around M=0, i.e. there are
usually lots of genes which are not differentially expressed,
but do not all sit exactly on M=0.

If you want to narrow down the list a bit, you could try using
limma from the command-line (using the function classifyTestsF),
to ask questions like "Which genes are differentially expressed
in one comparison, but are not differentially expressed in
another comparison?".

Hope this helps,
James