[BioC] mean vs median of pixel values within spot mask

[Gordon K Smyth]

> Date: Fri, 25 Feb 2005 17:16:43 -0500
> From: Pita <pwilkinson_m at xbioinformatics.org>
> Subject: mean vs median of pixel values within spot mask wasRe: [BioC]
> 	something weird with read.maimages and .gpr files
> To: bioconductor at stat.math.ethz.ch
> Cc: bioconductor at stat.math.ethz.ch
>
> It seems that I answered my own question ....
>
> There was a typo that was entered into the Genepix software so the output
> would not parse. Limmas read.maimages looks specifically for the F532 Mean
> for the R channel, instead F594 Mean was written
>
> Since it is the case that the Mean of the pixels in the pot mask are used
> in limma as default: are they any arguments for using the mean values over
> the median values?

Limma uses the mean of the foreground pixels according to the intuitive argument given in

Smyth, G. K., Yang, Y.-H., Speed, T. P. (2003). Statistical issues in microarray data analysis.
Methods in Molecular Biology 224, 111-136. http://www.statsci.org/pubs/mareview.pdf

i.e., because the mean is proportion to the total number of molecules binding.

Limma uses the median of the background because (i) here the aim is to measure the ambient level,
not the total binding, and (ii) because it is desirable for noise reduction for the background to
be slightly lower than foreground, and the median of a right skew distribution is lower than the
mean.

> Would the assumption here be that the spot morphology
> must be impeccable?

I have not yet seen any compelling evidence that the median foreground is more reliable or more
tolerant of poor spot morphology than the median foreground in this context.  If you can cite such
evidence, it would a useful contribution to the debate.

Gordon

> Peter