[BioC] limma normexp background correction bug?

[Gordon K Smyth]

Dear James and Tobias,

Well, I have to eat some of the words from my previous post.  On further 
checking, I can see that backgroundCorrect() accesses all the different 
normexp methods correctly when the data object RG is a matrix, but not 
when it is an RGList.  When RG is an RGList, the normexp always uses the 
default (saddlepoint) parameter estimates, regardless of which 
normexp.method option was input.

As usual, James is correct, and this also confirms part of what Tobias 
said.

The saddlepoint approximation has always been the default normexp method, 
so hopefully the current behaviour has not given anyone unexpected 
results.

I've fixed the bug today.

Best wishes
Gordon

> Date: Mon, 26 Oct 2009 15:16:44 -0400
> From: "James W. MacDonald" <jmacdon at med.umich.edu>
> Subject: Re: [BioC] limma normexp background correction bug?
> To: Tobias Straub <tstraub at med.uni-muenchen.de>
> Cc: bioc list <bioconductor at stat.math.ethz.ch>
>
>
> Tobias Straub wrote:
>> hi
>>
>> i was playing around with various background correction methods using
>> limma and discovered that whatever normexp.method i call for the
>> "normexp" method, the result is exactly the same.

...

> So normexp *is* used, but it appears that if you pass in either method =
> "rma" or method = "normexp", you will always end up using method =
> "saddle" for normexp.fit().
>
> Whether or not this is intentional or a bug is something only Gordon can
> decide.
>
> Best,
> Jim
>>
>> best
>> T.

...

>> ----------------------------------------------------------------------
>> Dr. Tobias Straub ++4989218075439 Adolf-Butenandt-Institute, M?nchen D
>>
> -- 
> James W. MacDonald, M.S.
> Biostatistician
> Douglas Lab
> University of Michigan
> Department of Human Genetics
> 5912 Buhl
> 1241 E. Catherine St.
> Ann Arbor MI 48109-5618
> 734-615-7826