[BioC] Problem with duplicateCorrelation in LIMMA

[Gordon K Smyth]

Dear York,

Thank you for sending me your data separately.  If you do a summary of 
your data

   summary(M)

you will see the problem immediately.  The 4th column of your data 
contains numbers which are +- 10^11 whereas the first 3 columns contain 
numbers which are close to zero.  There has been an error in the 
generation of your data.

The morale is that you need to look at your data and assess its quality 
before throwing it into an estimation procedure.

Best wishes
Gordon

> Date: Tue, 19 May 2009 13:40:58 -0700
> From: York Lee <yorklee70 at gmail.com>
> Subject: Re: [BioC] Problem with duplicateCorrelation in LIMMA
> To: Bioconductor at stat.math.ethz.ch
>
> I have four columns of log2-transformed ratio data
> (http://ifile.it/ija65rq). I want to analyze it with LIMMA, but stuck
> with duplicateCorrelation:
>
>> library(limma)
>> M <- read.delim('M.txt')
>> design <- cbind(c(-1, 0, -1, 0), c(0, -1, 0, -1))
>> corr <- duplicateCorrelation(M, design=design, spacing=5776, ndups=3)
> Error in glmgam.fit(dx, dy, start = start, tol = tol, maxit = maxit,
> trace = trace) :
> ??????? Starting values give negative fitted values
>
> Does anybody has similar experience and any hints to share?
>
> York