[BioC] limma all adj.P.Val the same

[James W. MacDonald]

Hi Sam,

On 5/21/2012 12:12 PM, Sam McInturf wrote:
> Hello Everyone,
>     I am working on expression profiling of wheat, where five tissue
> types were analyzed at anthesis (flowering) and 14 days after anthesis
> (DAA14).  Because of the amount of development that occurs during the
> first two weeks of grain filling, we expect to see differentially
> expressed genes (Code below). I approached this as a 5x2 factorial
> problem and took my lead from the limma guide (ch 8.7).  When calling
> topTable for any contrast, the list of P.Values contains a range of
> values (small small, some large).  While the adj.P.Val column contains
> the same (large) value for every gene reported ( up to n = 10,000).
>
> The adj.P.Vals that comeback for each contrasts range from 0.87 to 0.9999..
>
> Note the "ABSURD" contrast, when comparing different tissue types, we
> expect a lot of DE genes, but still no DE reported
>
> I subsetted my data to only analyze one tissue across the two times,
> and then doing the analysis as below, and if it is approached it as a
> 'comparison of two groups', and still was returned adjusted p values
> that were all the same, high value.
>
> Any idea why my q-values are all the same?

Without any more information it is difficult to say. The stock answer 
would be that you don't have enough power to detect any differences. I 
have seen this on occasion, especially when there isn't much replication.

You could help us by giving your targets file and an example of one of 
your topTables.



>
> ################################################
> library(limma)
> library(affy)
> #read in our data
> target<- readTargets("--------/TargetCEL.txt")
> data<- ReadAffy(filenames = target$FileName)
>
> eset<- rma(data)
> #Building our model.  Let us model this as a 5x2 factorial problem
> (five tissues, two times)
> factors<- paste(target$Timing, target$Sample, sep = ".")
> factors<- factor(factors, levels = c("ANTH.SPIKE",  "ANTH.PEDUNCLE",
> "ANTH.STEM", "ANTH.FLAGLEAF",  "ANTH.NODES", "DAA14.SPIKE",
> "DAA14.PEDUNCLE", "DAA14.STEM", "DAA14.FLAGLEAF", "DAA14.NODES"))
> design<- model.matrix(~0+factors)
> colnames(design)<- levels(factors)
> contrastMat.TissueWise<- makeContrasts(
>                   Spike = DAA14.SPIKE - ANTH.SPIKE,
>                   PEDUNCLE = DAA14.PEDUNCLE - ANTH.PEDUNCLE,
>                   STEM = DAA14.STEM - ANTH.STEM,
>                   FLAGLEAF = DAA14.FLAGLEAF - ANTH.FLAGLEAF,
>                   NODES = DAA14.NODES - ANTH.NODES,
>                   ABSURD = ANTH.STEM - DAA14.SPIKE,
>                    levels = design
>                   )
>
> fit3<- lmFit(eset, design)
> fit2<- contrasts.fit(fit3, contrastMat.TissueWise)
> fit<- eBayes(fit2)
>
> topTable(fit2, coef = ___, n = 100)

I am assuming this is a typo (fit2 rather than fit)?

Best,

Jim


> --
> Sam McInturf
>
> _______________________________________________
> 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