> Date: Wed, 23 Oct 2013 08:03:19 +0100
> From: Michael Breen <breenbioinformatics at="" gmail.com="">
> To: Ryan <rct at="" thompsonclan.org="">
> Cc: Bioconductor Mailing List <bioconductor at="" stat.math.ethz.ch="">,
> "bioconductor at r-project.org" <bioconductor at="" r-project.org="">
> Subject: Re: [BioC] ANCOVA microarray time-course continuous &
> categorical variables / Limma extensions
>
> I have done this in the past, although I did not have much faith in
it,
> maybe because of the way I create the design matrix.
>
> For example I do it this way:
>
> Treat <- factor(paste(targets$Condition,targets$Time,sep="."))
> design <- model.matrix(~0+Treat)
> colnames(design) <- levels(Treat)
> colnames(design)
> [1] "Case.Time2" "Case.Time1" "Control.Time2" "Control.Time1"
>
>
> #introduce continuous variable, add to design
> CES <- factor(targets$CES)
Here is the problem. You are declaring your continuous variable to be
categorical, which is presumably not want you want to do.
In R, a "factor" means a categorical variable. Anything that is not a
factor is considered to be continuous.
You could either remove the factor() command here, or else simply use
design <- model.matrix(~0+Treat+CES)
Best wishes
Gordon
> design2 <- cbind(design, CES)
> colnames(design2)
> [1] "Case.Time2" "Case.Time1" "Control.Time2" "Control.Time1"
> "CES"
>
> I go on to duplicate correlation and lmfit then make these contrasts
to
> substract away the effect of the continuous variable from our
observations.
>
> cm <- makeContrasts(
> CaseEffect = Case.Time2-Case.Time1,
> ControlEffect = Control.Time2-Control.Time1,
> ContrastEffect =
(Case.Time2-Case.Time1)-(Control.Time2-Control.Time1),
> CaseEffectNoCov = (Case.Time2-Case.Time1)-(CES),
> ControlEffectNoCov = (Case.Time2-Case.Time1)-(CES),
> ContrastEffectNoCov = (Case.Time2-Case.Time1)-(CES) -
> (Case.Time2-Time1)-(CES),
> levels=designnew)
>
> etc...
>
> Our results are 'different' so it would be useful to have another
opinion
> regarding this set-up.
>
> Michael
>
>
>
>
>
>
>
>
> On Tue, Oct 22, 2013 at 11:03 PM, Ryan <rct at="" thompsonclan.org="">
wrote:
>
>> Including continuous covariates in design matrices is R is just as
easy as
>> including categorical ones. Instead of creating a column for each
degree of
>> freedom in the categorical variable, you just end up with a single
column
>> that simply contains the values of the continuous variable. Try
using the
>> model.matrix function with a combination of your categorical
variables and
>> continuous ones to see what it does.
>>
>> -Ryan
>>
>>
>> On Tue Oct 22 14:36:38 2013, Richard Friedman wrote:
>>
>>>
>>> On Oct 22, 2013, at 5:30 PM, Michael Breen wrote:
>>>
>>> Hi all,
>>>>
>>>> Our lab analyzes gene-expression from microarray and RNAseq
platforms.
>>>> Currently, I am looking for a package to test differential
expression
>>>> (DE)
>>>> while considering continuous variables that may alter gene-
expression
>>>> profiles. In other words, an ANCOVA type tool. I am quite
familiar with
>>>> Limma (ANOVA) but including continuous variables is not very well
>>>> described.
>>>>
>>>> Specifically, we have a project were two groups can be modeled
over the
>>>> same 2 time points. One group starts healthy and ends in a
disease state.
>>>> The other group starts healthy and remains healthy.
>>>>
>>>> We are interested in identifying genes uniquely responding within
one
>>>> group
>>>> and not in the other. Thus, we have implemented a longitudinal
contrast
>>>> with linear modeling through Limma. However, we are also
interested in
>>>> adding one or two continuous variables (tumor size, time spent
>>>> meditating,
>>>> the amount of drinks one consumes etc..) to check if gene
expression
>>>> differences or similarities may be due to these factors instead
of due to
>>>> belonging to a certain class. Limma seems to test categorical
variables,
>>>> but I don't think it is capable of either correlating gene-
expression to
>>>> continuous variables.
>>>>
>>>> If not, can someone recommend a tool which may be appropriate
for such a
>>>> situation?
>>>>
>>>> Yours,
>>>>
>>>> Michael
>>>>
>>> Dear Michael and list,
>>>
>>> I think that you write the design and contrast matrices
>>> exactly as you would for an ANCOVA in R only you do the
>>> fit and Bayesian correction in Limma.
>>>
>>> Perhaps someone who has had experience doing this
>>> kind of analysis can comment.
>>>
>>> Best wishes,
>>> Rich
>>>
>>> Richard A. Friedman, PhD
>>> Associate Research Scientist,
>>> Biomedical Informatics Shared Resource
>>> Herbert Irving Comprehensive Cancer Center (HICCC)
>>> Lecturer,
>>> Department of Biomedical Informatics (DBMI)
>>> Educational Coordinator,
>>> Center for Computational Biology and Bioinformatics (C2B2)/
>>> National Center for Multiscale Analysis of Genomic Networks
(MAGNet)/
>>> Columbia Department of Systems Biology
>>> Room 824
>>> Irving Cancer Research Center
>>> Columbia University
>>> 1130 St. Nicholas Ave
>>> New York, NY 10032
>>> (212)851-4765 (voice)
>>> friedman at cancercenter.**columbia.edu <friedman at="" cancercenter.columbia.edu="">
>>> http://cancercenter.columbia.**edu/~friedman/<http: cancercenter.="" columbia.edu="" ~friedman=""/>
>>>
>>> In memoriam, Frederik Pohl
______________________________________________________________________
The information in this email is confidential and
intend...{{dropped:4}}
