[BioC] limma subsetting issue

[Gordon K Smyth]

Dear Ariel,

Thanks for pointing out this problem.  In limma 2.7.12 (on Bioconductor devel now) the design
matrix is no longer subsetted when an MArrayLM is subsetted.

Best wishes
Gordon

> Date: Sat, 02 Sep 2006 09:13:24 -0300
> From: Ariel Chernomoretz <ariel at df.uba.ar>
> Subject: [BioC] limma subsetting issue
> To: "'Bioconductor'" <bioconductor at stat.math.ethz.ch>
> Message-ID: <44F97564.7080007 at df.uba.ar>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
> Dear Gordon and BioC users,
>
> I have recently updated R and BioC packages (R.2.3.0 and limma 2.7.9)
> and found the following issue
> when I try to subset a MArrayLM object.
>
> "f" is a MArrayLM object obtained using 'contrasts.fit' and
>
>  > class(f)
> [1] "MArrayLM"
> attr(,"package")
> [1] "limma"
>
>  > dim(f$contrasts)  #lot of contrasts
> [1]  8 21
>
>  > dim(f$coef)
> [1] 10984    21
>
>  > class(f[,c(1,4,8)])  #subsetting columns 1:8 runs smoothly
> [1] "MArrayLM"
> attr(,"package")
> [1] "limma"
>
>  > class(f[,9])    #subsetting any column between 9 - 21 does not work
> Error in "[.MArrayLM"(f, , 9) : subscript out of bounds
>
> I think the problem relies on how subsetting works on the design matrix
> of the MArrayLM object (which in my case has 8 columns)
> Under R.2.21 and limma 2.4.7 this was not an issue (almost).
> The following warning was raised:   "Warning message:  design matrix is
> singular"
> but  the subsetting operation was performed anyway.
>
> I know that from the point of view of the MArrayLM object integrity this
> is not the ideal behavior, but
> it proved to be really usefull (at least for me). If this new behavior
> is limma's responsability, and not R's, may I suggest to
> change things in orther to allow subsetting using all contrasts again?
> Thanks
> Best
> Ariel./