[BioC] limma: design for paired data with multiple time points

[James W. MacDonald]

Hi Jonathan,

On 5/19/2014 9:28 PM, Jonathan Ellis wrote:
> Dear list,
>
> I am analysing some microarrays with the limma package, and so far have
> discovered no significant DEGs.  I'm unsure if this is because there
> really are no DEGs or I've misunderstood the limma package.  I'm hoping
> someone with more experience can tell me if my analysis is correct, or,
> if not, where I've gone wrong.
>
> I have arrays collected at three time points (t0, t14 and t56) from the
> same patients, so I have a data frame that resembles:
>
> Patient Time
> 1       t0
> 1       t14
> 1       t56
> 2       t0
> 2       t14
> 2       t56
> etc.
>
> and I'm interested in comparing t14 to t0, t56 to t0 and t56 to t14,
> whilst accounting for fact that array are from the same patient.  My
> analysis has been:
>
> design <- model.matrix(~ 0 + Time + Patient)
> colnames(design) <- c('t0', 't14', 't56', 'p1', 'p2', 'p3', 'p4', 'p5', 'p6')
> array.weights <- arrayWeights(x.filtered, design)
> fitw <- lmFit(x.filtered, design, weights = array.weights)
> contrast.matrix <- makeContrasts(t14-t0, t56-t14, t56-t0, levels = design)
> fit2 <- contrasts.fit(fitw, contrast.matrix)
> fit2 <- eBayes(fit2)
>
> I would be very grateful if someone can tell me if this analysis is
> correct or not.

Looks OK to me. When you say 'no significant DEGs', what exactly do you 
mean? You can certainly use a relatively large FDR, if you are willing 
to accept failures when you validate.

Best,

Jim


>
> Cheers,
> Jonathan
>
> _______________________________________________
> Bioconductor mailing list
> 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