[BioC] Zero variability in RMA

[James MacDonald]

This is an artifact of the medianpolish algorithm. If you are using 3 or
5 chips you will get some expression values that are identical for all
chips (much worse with 3 chips). You can see why this is if you take the
quantile normalized pm values from one of these genes and run
medpolish() on them (note here that they have to be normalized!).

What happens is that the overall median is subtracted out, then the row
median, but at this point the column median for all columns is zero.
Since the expression values are the overall median + column median, the
expression values are identical for all chips.

I don't think there is anything you could/should do with these data.
Long story short, these genes are likely not changing expression so they
are probably not interesting anyway. If you are having problems with
zeros in your denominators, you could simply remove the 486 genes and
then proceed. 

HTH,

Jim



James W. MacDonald
Affymetrix and cDNA Microarray Core
University of Michigan Cancer Center
1500 E. Medical Center Drive
7410 CCGC
Ann Arbor MI 48109
734-647-5623

>>> Angelo Canty <canty at math.mcmaster.ca> 06/07/04 02:06PM >>>
Hi,

I am analyzing a small microarray experiment with 5 RAE230_2 chips
with material from two strains of rat.

I loaded in the data using ReadAffy and then run the rma function to
normalize the data and produce the expression values.  In 486 (out of
31099) probesets, the expression values are identical over all 5
chips and in two others there was no variability within strain 
although there was across strains.  In either of these cases, I get
zero variability for the mean difference.  

I have examined the original probe level data as read from the CEL
files and there is variability across the chips in both the PM and
MM measurements so my conclusion is that this is an effect of the
rma or normalization steps.  Has anyone else witnessed this effect?
What is the preferred way to deal with this lack of variability?

Thanks for any help that you can provide.
Angelo

-- 
------------------------------------------------------------------
|   Angelo J. Canty                Email: cantya at mcmaster.ca     |
|   Mathematics and Statistics     Phone: (905) 525-9140 x 27079 |
|   McMaster University            Fax  : (905) 522-0935         |
|   1280 Main St. W.                                             |
|   Hamilton ON L8S 4K1                                          |

_______________________________________________
Bioconductor mailing list
Bioconductor at stat.math.ethz.ch 
https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor