[BioC] limma background correction

[Gordon K Smyth]

Dear Ulrike and Henrique,

The problem is more likely to be with the limma normexp.fit() function 
than with optim().

There is a rare numerical instability which causes an error like this with 
normexp background correction in limma 2.14.X.  Jeremy Silver and I did a 
lot of work on this late last year, and we think we have eliminated the 
problem.  The new (faster) code is in limma on the Bioconductor 
developmental repository.  You can try it out if you install R 2.8.0.  I'd 
be interested to know if it solves the problem for your data sets.

Best wishes
Gordon

> Date: Thu, 10 Jul 2008 17:35:51 +0200
> From: Ulrike Goebel <ugoebel at mpiz-koeln.mpg.de>
> Subject: Re: [BioC] limma background correction
> To: bioconductor at stat.math.ethz.ch
> Message-ID: <200807101735.52255.ugoebel at mpiz-koeln.mpg.de>
> Content-Type: text/plain;  charset="iso-8859-1"
>
> Hi Henrique,
>
> I recently got the same error with a Nimblegen dataset. Because in this
> dataset the background is actually not provided (the corresponding column is
> all-zero), I figured that was the reason ... However, this error occurs only
> with R-2.7.0, not with 2.6 or 2.5. So something must have changed in the
> optim() function (this is where it crashes in my case).
>
> Glad to hear that I am not the only one who stumbled across this !
>
> Ulrike
>
> On Thursday 10 July 2008 17:13, Henrique Proen?a wrote:
>> Hi,
>>
>> I am trying to analyze some quantarray files using limma Guide, when
>> suddenly it gives me the following error at the background normalization
>> step (using normexp method and offset = 50)
>>
>> Error in if (all(abs(delta) < 1e-10)) break :
>>    missing value where TRUE/FALSE needed
>>
>> I can't find why this happens.
>> Does anyone an idea of what is happening?
>
> --
>  Dr. Ulrike Goebel
>  Bioinformatics Support
>  Max-Planck Institute for Plant Breeding Research
>  Carl-von-Linne Weg 10
>  50829 Cologne
>  Germany
>  +49(0) 221 5062 121