[BioC] detailed analysis process of \"Comparisons Both Between and Within Subjects\" of edgeR

[Ryan C. Thompson]

Hello,

(Sorry about the incomplete message before. I tripped and fell on the 
send button.)

I think you might need to share what your targets file and your design 
matrix look like. If I had to guess, I would guess that your design 
matrix has too many columns for some reason (specifically, more columns 
than degrees of freedom). If you show us your design matrix, we could 
verify whether this is indeed the case, or whether there is some other 
problem.

-Ryan Thompson

On 01/08/2013 04:39 PM, Huajing Teng [guest] wrote:
> Dear Dr.Smyth,
> I am very interested  in "Comparisons Both Between and Within Subjects" of edgeR. Because one of my study is according to this design. But i found some detailed analysis have not been provided in the manual (After estimating the dispersions (code not shown), we can fit a linear model:), and i met some difficulty in using this pipline.  When i used y <- estimateGLMCommonDisp(y, design, verbose=TRUE), I got an error:
> error in solve.default(R, t(beta)) : Lapack routine dgesv: system is exactly singular
>
> I wonder whether you can provide some detailed analysis process (especially in data reading, dispersion evaluation ). Thanks for your kind assistance.
>
>   -- output of sessionInfo():
>
> source("http://bioconductor.org/biocLite.R")
> biocLite("edgeR")    biocLite("limma")
> library(edgeR)
>> targets <- read.delim("F:/teng/thyroidcarcinoma/gene_exp.diff/edgeR/2nd/geneexpression/allnosui/design.txt", check.names=FALSE, stringsAsFactors=FALSE)
>> targets
>> Patient <- gl(5,2,length=18)
>> Sex <- factor(targets$Sex, levels=c("M","F"))
>> Treatment <- factor(targets$Treatment, levels=c("N","T"))
>> data.frame(Sex,Patient,Treatment)
>> y <- read.delim("F:/teng/thyroidcarcinoma/gene_exp.diff/edgeR/2nd/geneexpression/allnosui/genecount.txt", row.names=1, header=T, stringsAsFactors=FALSE)
>> design <- model.matrix(~Sex+Sex:Patient+Sex:Treatment)
>> rownames(design) <- colnames(y)
>> y <- estimateGLMCommonDisp(y, design, verbose=TRUE)
> #error:
> #error in solve.default(R, t(beta)) : Lapack routine dgesv: system is exactly singular
>
> --
> Sent via the guest posting facility at bioconductor.org.
>
> _______________________________________________
> 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