[BioC] Limma Modeling

[Sean Davis]

The other way to think of this problem is as a classification problem.  If
you have two groups, those that gain weight and those that do not, you can
use tools like SVM or randomForests to determine those genes that are most
predictive of weight gain.  As Naomi points out, there may be zero to MANY
genes that can classify your samples correctly.  At least for randomForests,
the question can be framed as a regression, as well as a classification
problem.

Sean


On 7/20/06 11:19, "Naomi Altman" <naomi at stat.psu.edu> wrote:

> The problem is that you have thousands of genes which you have
> summarized as "gene expression".  This means that you can obtain a
> perfect fit with many different sets of genes.
> 
> So, usually you need to do some gene selection before you fit the
> model.  Then you could do ordinary variable selection as in ordinary
> linear regression, or possibly use a method like ridge regression.
> 
> --Naomi
> 
> At 12:00 PM 7/20/2006, Katarzyna Bryc wrote:
>> Dear List,
>> 
>> I have a question on correctly modeling a situtation to find
>> significantly differentiated genes with Limma. I have Affy arrays for
>> pediatric patients collected before the patients were treated with a
>> drug for 4 months. After this time period, some patients had a side
>> effect of significant weight gain, while others did not. I wish to find
>> the genes which significantly differentiate patients who gained weight
>> with the treatment from those who did not suffer this side effect. Since
>> these are pediatric patients, I also wish to control for Sex and Age
>> (continuous variable).
>> 
>> I understand that with Limma I can model the following:
>> 
>> Gene Expression = Age + Sex + Weight Gain
>> 
>> but I actually wish to look at Weight Gain as the dependent variable,
>> and Gene Expression as one of the independent variables (I still control
>> for Age and Sex). Thus, I actually wish to model
>> 
>> Weight Gain = Age + Sex + Gene Expression
>> 
>> My questions are:
>> 1. Will these two models give me the same results for finding genes
>> significant in predicting Weight Gain?
>> 2. If not, is there a way to model this using either Limma or another
>> Bioconductor method?
>> 
>> Thank you for any helpful words,
>> Kasia Bryc
>> 
>> _______________________________________________
>> Bioconductor mailing list
>> Bioconductor at stat.math.ethz.ch
>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>> Search the archives:
>> http://news.gmane.org/gmane.science.biology.informatics.conductor
> 
> 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
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives:
> http://news.gmane.org/gmane.science.biology.informatics.conductor