[BioC] Limma-include interaction term
James W. MacDonald
jmacdon at uw.edu
Tue Jun 12 21:14:44 CEST 2012
Hi Jack,
On 6/12/2012 2:22 PM, Yao Chen wrote:
> I am confused. If I want to get differential expressed genes between
> treatment and control with age interaction. Which one should I use
> :topTable (fit2, coef=2) or topTable (fit2, coef=4) .
I have no idea what you are asking, so I will just try to restate things.
If you fit a model with treatment and age, then any genes with a
significant treatment - control contrast implies that after adjusting
for age of the subjects, there is a difference in expression of that
gene between treated and control subjects. One way to conceptualize this
model is that you are fitting two lines to your data, one for treated
and one for control, but you are constraining these lines to have the
same slope (e.g., they are parallel).
If you fit a model that includes a treatment:age interaction, you are
doing the same exact thing, only now you are fitting two lines and
allowing the slopes to vary as well. So the example I gave you, coef 2
tells you which genes are differentially expressed between treatment and
control, after adjusting for age of the subjects and allowing the slopes
to vary. Coef 4 tells you whether or not the slopes are different for
the treated and control subjects.
The reason coef 4 might be interesting is because it will allow you to
find genes that (as an example) are expressed at the same level in
untreated subjects regardless of age, but in treated subjects the
expression increases dramatically as a function of age.
Best,
Jim
>
> Thanks,
>
> Jack
>
> 2012/6/12 James W. MacDonald <jmacdon at uw.edu <mailto:jmacdon at uw.edu>>
>
> Hi Jack,
>
>
> On 6/12/2012 1:54 PM, Yao Chen wrote:
>
> Thanks James. That's exactly what I want to know.
>
> But I am not sure I fully understand the differential
> expressed genes in topTable. For (fit2, coef=2), did I get the
> genes without considering treat:age interaction, as my
> previous design matrix . And (fit2, coef=4) gives me the genes
> considering treat:age interation.
>
>
> No. When you fit a model with a bunch of coefficients, a given
> coefficient measures the marginal effect of the coefficient after
> accounting for all other coefficients in the model.
>
> In conventional linear modeling (where you aren't fitting
> thousands of models at once), you would probably fit a model with
> and without the interaction term and then test to see if the
> interaction term is significant. This is difficult to do in the
> context of a microarray analysis, so people generally just throw a
> bunch of coefficients in a model and look for significant genes.
>
> If you then wanted to do some other tests with a subset of your
> genes I suppose you could, but people generally pick 'interesting'
> genes and go to functional studies.
>
> Best,
>
> Jim
>
>
>
> Jack
>
> 2012/6/12 James W. MacDonald <jmacdon at uw.edu
> <mailto:jmacdon at uw.edu> <mailto:jmacdon at uw.edu
> <mailto:jmacdon at uw.edu>>>
>
> Hi Jack,
>
> The conventional method is to use the model.matrix()
> function. I
> have no idea what your data look like, so here is a random
> example:
>
> > treat <- factor(rep(0:1, each = 5))
> > treat
> [1] 0 0 0 0 0 1 1 1 1 1
> Levels: 0 1
> > age <- sample(25:35, 10, TRUE)
> > age
> [1] 32 30 32 35 29 26 27 25 33 34
> > model.matrix(~treat*age)
> (Intercept) treat1 age treat1:age
> 1 1 0 32 0
> 2 1 0 30 0
> 3 1 0 32 0
> 4 1 0 35 0
> 5 1 0 29 0
> 6 1 1 26 26
> 7 1 1 27 27
> 8 1 1 25 25
> 9 1 1 33 33
> 10 1 1 34 34
> attr(,"assign")
> [1] 0 1 2 3
> attr(,"contrasts")
> attr(,"contrasts")$treat
> [1] "contr.treatment"
>
> Note that this uses a different parameterization. In this
> case the
> treat1 coefficient is the difference between the treated and
> untreated samples (so you wouldn't specify a
> contrasts.matrix, you
> just do lmFit() and then eBayes()). The treat1:age coefficient
> captures the difference between the slopes for the treated and
> untreated samples.
>
> So topTable(fit2, coef=2) gives you genes that are
> differentially
> expressed between treated and untreated and topTable(fit2,
> coef=4)
> gives you genes where the change in expression at different
> ages
> varies between treated and untreated subjects.
>
> Best,
>
> Jim
>
>
>
>
>
> On 6/12/2012 12:57 PM, Yao Chen wrote:
>
> Thanks, James
>
> How to include "age:treatment" interaction in the
> design matrix?
>
> Jack
>
> 2012/6/12 James W. MacDonald <jmacdon at uw.edu
> <mailto:jmacdon at uw.edu>
> <mailto:jmacdon at uw.edu <mailto:jmacdon at uw.edu>>
> <mailto:jmacdon at uw.edu <mailto:jmacdon at uw.edu>
>
> <mailto:jmacdon at uw.edu <mailto:jmacdon at uw.edu>>>>
>
>
> Hi Jack,
>
>
> On 6/12/2012 10:10 AM, Yao Chen wrote:
>
> Dear All,
>
> I try to find differential expressed genes between
> treat and
> untreated
> samples, and also I want to include the age effects.
>
> The design matrix is like this:
>
> treat untreated age
> 1 0 30
> 0 1 40
> 1 0 35
>
>
> The "treat" is factor, but "age" is continuous. How
> can I set the
> "cont.matrix"?
>
>
> Pretty much just like you (or at least I) would expect:
>
> contrast <- makeContrasts(treat - untreat, levels =
> design)
>
> But note that the design you are specifying allows
> different
> intercepts, but the slope is assumed to be the same
> for treated
> and untreated. If you want to allow different slopes as
> well, you
> need to introduce an age:treatment interaction term.
> Here I am
> assuming you have more than three samples.
>
> Best,
>
> Jim
>
>
>
> Thanks,
>
> Jack
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org <mailto:Bioconductor at r-project.org>
> <mailto:Bioconductor at r-project.org
> <mailto:Bioconductor at r-project.org>>
> <mailto:Bioconductor at r-project.org
> <mailto:Bioconductor at r-project.org>
>
> <mailto:Bioconductor at r-project.org
> <mailto:Bioconductor at r-project.org>>>
>
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives:
> http://news.gmane.org/gmane.science.biology.informatics.conductor
>
>
> -- James W. MacDonald, M.S.
> Biostatistician
> University of Washington
> Environmental and Occupational Health Sciences
> 4225 Roosevelt Way NE, # 100
> Seattle WA 98105-6099
>
>
>
> -- James W. MacDonald, M.S.
> Biostatistician
> University of Washington
> Environmental and Occupational Health Sciences
> 4225 Roosevelt Way NE, # 100
> Seattle WA 98105-6099
>
>
>
> --
> James W. MacDonald, M.S.
> Biostatistician
> University of Washington
> Environmental and Occupational Health Sciences
> 4225 Roosevelt Way NE, # 100
> Seattle WA 98105-6099
>
>
--
James W. MacDonald, M.S.
Biostatistician
University of Washington
Environmental and Occupational Health Sciences
4225 Roosevelt Way NE, # 100
Seattle WA 98105-6099
More information about the Bioconductor
mailing list