[BioC] Can subject be treated as fixed effect in linear model with microarray data
Guilherme J. M. Rosa
grosa at wisc.edu
Wed Jun 6 17:59:38 CEST 2007
If I understood right your design, you had both 'normal' and 'cancer'
tissues sampled in each subject, right?
So in this case you do not have 'tissue nested in subject', like you say.
Subjects are within gender, but crossed with tissues.
You have, as you say, a 2-way anova (factors: Gender and tissue), but you
have repeated measurements on each subject, i.e. the two tissues assayed in
ach subject. In this case you can see subjects as a blocking factor, but
actually you have two sampling levels, like a split-plot design.
I would suggest a model like:
y = mu + gender + subject(gender) + tissue + gender*tissue + e
where gender, tissue and the interaction gender*tissue are considered as
fixed effects; and subject within gender is random (which is actually the
error term to compare the two genders).
You may want to use the MAANOVA software to run such a model
LIMMA may not give you the flexibility to implement it.
Guilherme J. M. Rosa
Department of Dairy Science
University of Wisconsin - Madison
460 Animal Science Building
1675 Observatory Dr.
Madison, WI 53706 USA
Phone: + 1 (608) 265-8617
Fax: + 1 (608) 263-9412
E-mail: grosa at wisc.edu
----- Original Message -----
From: "Naomi Altman" <naomi at stat.psu.edu>
To: "shirley zhang" <shirley0818 at gmail.com>;
<Bioconductor at stat.math.ethz.ch>
Sent: Wednesday, June 06, 2007 9:28 AM
Subject: Re: [BioC] Can subject be treated as fixed effect in linear model
with microarray data
> Model 3 is completely illegal. Model 2 is sometimes used when there
> are few within subject observations (as here). However, I would not
> do that here. I would use an eBayes method such as limma to improve
> At 11:34 AM 6/5/2007, shirley zhang wrote:
>>In a microarray data, there are 20 subjects grouped by Gender, each
>>subject has 2 tissues (normal vs. cancer).
>>In fact, it is a 2-way anova (factors: Gender and tissue) with tissue
>>nested in subject. I've tried the following:
>>Model 1: lme(response ~ tissue*Gender, random = ~1|subject)
>>Model 2: response ~ tissue*Gender + subject
>>Model 3: response ~ tissue*Gender
>>It seems like Model 1 is the correct one since my experiment design is
>>nested design. However, I got a few significant genes for Gender
>>effect from Model 1 so I want to use Model 2 or Model 3. Can anybody
>>tell me whether Model2 is
>>Bioconductor mailing list
>>Bioconductor at stat.math.ethz.ch
>>Search the archives:
> Naomi S. Altman 814-865-3791 (voice)
> Associate Professor
> Dept. of Statistics 814-863-7114 (fax)
> Penn State University 814-865-1348 (Statistics)
> University Park, PA 16802-2111
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> Search the archives:
More information about the Bioconductor