[BioC] limma control spot weights and normalization

[Gordon K Smyth]

> Date: Mon, 23 Jan 2006 14:02:03 -0800
> From: "Kathryn V. Steiger" <ksteiger at illumigen.com>
> Subject: [BioC] limma control spot weights and normalization
> To: <bioconductor at stat.math.ethz.ch>
>
> Will setting control spots to zero affect normalization within/among
> arrays?

Yes.  normalizeWithinArrays() uses weights if supplied.

> I've inherited a set of scripts for gene expression analysis of Agilent
> two-color arrays.  We are using ScanExpress software by Perkins to get
> convert the images to R,G intensities

Do you mean ScanArray Express by PerkinElmer?

> so "read.maimages" and readGAL
> have been adapted.

read.maimages() is customisable, so it almost never necessary to re-write it, even for a
non-standard image analysis program.

> I just noticed that there is a line that sets the RG$genes$ControlType
> values to zero if they are positive or negative controls.  Then those
> are included in the weights file.
>
> Is this a problem?  Should I be using something more like
> wtIgnore.Filter?

It is very hard to give you any advice because the line you refer to is not part of the limma
package.  So we don't know who wrote the code or what they were trying to achieve exactly.  I can
only say that it often is a good idea to remove Agilent positive controls from the normalization. 
And, since you're not using QuantArray, it is unlikely that you have a column called
wtIgnore.Filter.

Best wishes
Gordon

> We have 1502 control spots / array.
>
> Thanks,
>
> Kathryn Steiger
> Illumigen Biosciences
> Seattle, WA