[BioC] high poportion of identical expression measure with RMA and 3 arrays

[Adaikalavan Ramasamy]

I was pondering about the issue of minimum number of required arrays for
RMA and GCRMA when I came across a thread by James MacDonald
http://files.protsuggest.org/biocond/html/3347.html. He mentioned that
median polish has a peculiar feature that returned identical values for
odd number of arrays, especially 3 arrays. I am not sure if this issue
has been addressed since but here is some empirical evidence to support
this in case anyone is interested.

### Codes ###

library(affy)
v <- rep(NA, 10)

for(i in 2:10){
  a <- exprs( justRMA( filenames=list.celfiles()[1:i] ))
  v[i] <- mean( apply(a, 1, var) == 0 )  # proportion identical
  gc(); print(i)
}
100*v


### Results ###

# Dataset 1 : hu6800 (west)
> 100 * v
 [1]         NA  0.0000000 13.1434984  0.0000000  0.1262449  0.0000000
 [7]  0.0000000  0.0000000  0.0000000  0.0000000

# Dataset 2 : hgu-95av2 (febbo)
> 100 * v
 [1] NA 0.007920792 7.912871287 0.000000000 0.095049505 0.000000000
 [7] 0.000000000 0.000000000 0.000000000 0.000000000

# Dataset 3 : HGU-133A (pga)
> 100 * v
 [1]         NA  1.5348023 16.0750348  0.0000000  0.3590181  0.0000000
 [7]  0.0000000  0.0000000  0.0000000  0.0000000


When there are 3 arrays, there is a high proportion of probesets with
identical values (i.e. zero variance). I am trying to see if the same
effect can be seen with justGCRMA.

misc : affy 1.4.32, gcrma 1.1.0, R-1.9.0 (12/04/2004) and FC 2.