[BioC] microarray analysis of a dose response * strain experiment

[Kimpel, Mark William]

Sean,

Perhaps it will have to be, I can think of two ways to do that and
neither seems entirely satisfactory. Firstly, one could assume that the
response (for responding genes) would be linearly related to the dose or
the log of the dose, but this might not be the case. So regressing by
dose in a linear model might not be correct for all or even most of the
affected genes. Secondly, one could simply assume that the dose is a
factor with 4 non-ordered levels and look at contrasts and interactions
for each level. This would be the approach I am most familiar with using
Limma. This would, however, seem to be throwing information away
regarding the relationship of the doses to one another.

Mark

Mark W. Kimpel MD 

 

(317) 490-5129 Work, & Mobile

 

(317) 663-0513 Home (no voice mail please)

1-(317)-536-2730 FAX

-----Original Message-----
From: Sean Davis [mailto:sdavis2 at mail.nih.gov] 
Sent: Saturday, November 04, 2006 1:39 PM
To: bioconductor at stat.math.ethz.ch
Cc: Kimpel, Mark William; McBride, William J.
Subject: Re: [BioC] microarray analysis of a dose response * strain
experiment

On Saturday 04 November 2006 13:20, Kimpel, Mark William wrote:
> My group is writing a grant with a proposed dose response experiment
on two
> different rat strains that I have been tasked to provide an analysis
method
> for. Briefly, we have two rat strains that have different preferences
for
> alcohol (one drinks, the other doesn't). We are going to give each
line
> injections for alcohol to see if gene expression in the brain is
> differentially affected between the 3 strains. We don't, however, know
> which of several possible doses of alcohol will provide the greatest
effect
> on each of the thousands of genes on our Affy chipset. So, we are
proposing
> to give each line one of 4 doses (zero, 0.5, 1.0, and 2.0 mg/kg). For
any
> gene, we have no way of knowing a priori what shape the dose response
curve
> will take. We are, for screening purposes, not really interested in
the
> shape of the curve, only that it is not a line with a slope of zero
(i.e.
> no response). We are also, for screening purposed, only interested to
know,
> for each gene, if the response of strain A is different from strain B.
In
> other words, what we want to know is the interaction between strain
and
> dose response.
>
> I have searched the literature and the Bioconductor mailing list and
cannot
> find a reference to an experiment of this sort. Can anyone provide
some
> advice?

This can't be handled with a linear model?

Sean