[BioC] Limma analysis

[Gordon Smyth]

At 02:25 PM 30/10/2003, Pete wrote:
>Sorry I didn't explain that particularly well, firstly how do I create a
>design matrix for this experiment?

I simply haven't been clever enough to figure out a way to automate the 
process of creating the design matrix for direct designs with two-colour 
arrays. Send me offline (1) your targets file (see the limma 1.3.0 manual 
for the meaning of a targets file) and (2) what comparisons you want to 
make between your samples or questions you want to answer, and I will try 
to get someone to suggest a design matrix.

>Also as far as I can see the read.imagene function doesn't read in the flag
>information for each file, in imagene each spot can have a flag value from
>0-8? and in this case we want to ignore completely anything which is non
>zero. Presumably this could be specified in the wt.fun argument, but i'm
>unsure precisely how to do this? I have tried to modify the wtflags function
>but without success.

mywtfun <- function(x) {
as.numeric(x[,"Flag"] == 0)
}

RG <- read.maimages(files, source="imagene", wt.fun=mywtfun)

Please note: I do not personally recommend ignoring flagged points in this 
way. I would personally down-weight them somewhat but would not ignore them 
completely. I don't think that spots split cleanly in this way into good 
and bad spots and I don't have anything like this sort of faith in 
Imagene's (or any other program's) ability to pick one from the other.

Regards
Gordon


>Cheers
>Pete
>
>
>----- Original Message -----
>From: "Gordon Smyth" <smyth at wehi.edu.au>
>To: "Pete" <p.underhill at har.mrc.ac.uk>
>Cc: <bioconductor at stat.math.ethz.ch>
>Sent: Tuesday, October 28, 2003 2:50 AM
>Subject: Re: [BioC] Limma analysis
>
>
> > At 10:04 AM 28/10/2003, Pete wrote:
> > >Hi all,
> > >I have been using limma now for a couple of weeks, and I think I have
>pretty
> > >much got the hang of most of it. However, now I want to analyse a
>slightly
> > >more complex experiment, can anyone give me some guidance as how to deal
> > >with this.
> > >      Firstly the experimental design is as follows, there are four
>samples
> > >wildtype tissue A, wildtype tissue B, mutant tissue A and mutant tissue
>B.
> > >Each sample has been compared to eachother in triplicate (inlcuding a dye
> > >swap, and one independant sample). To complicate things further an
> > >additional set of WT A v WT B was also done in triplicate using a
>different
> > >method.
> > >     The slides are 7.5k oligos spotted in duplicate ( the duplicates are
>in
> > >the same block 10 rows below the first copy), although there are control
> > >genes which appear more than twice on the arrays. My files are imagene
> > >output files where the cy5 and cy3 are contained in separate files.  Also
> > >the imagene output contains spots which are flagged and would need to be
> > >removed from the analysis (meaning that a particular gene could have none
> > >one or two measurements for it).
> > >
> > >What do you think the best strategy to deal with this design is?
> >
> > Well, everything in your experiment is straight down the line as far as
> > limma is concerned. You haven't really said what is is about this
> > experiment which you're not sure how to deal with. Is the problem the
> > design matrix or something else?
> >
> > Gordon
> >
> > >Cheers
> > >
> > >Pete