[BioC] affy expression values

[Laurent Gautier]

On Mon, Jul 29, 2002 at 08:39:30AM -0400, Larry wrote:
> Greetings Dr. Irizarry and Dr. Gautier,


...thanks... but I still cannot claim for the two letters to preceede my
name... ;)

> 
> I have a question regarding the affy procedures that I am hoping you can help me with.   I am in the process of exploring use of affy analysis as an alternative to using the Affymetrix expression values and in doing so I am trying to understand the process for generating expression values. Specifically, the medianpolish default procedure used by the express function.  
> 
> As I understand it, in calculating expression values for multiple chips,  medianpolish does not treat the chips independently. It runs through this process of removing row and column medians for all the chips then generates an overall value for all the chips with corrections for each chip.  The expression value generated is then calculated from these two values.  Does this then mean that I should expect different results depending on the number of chips I am analyzing? 
> 


technically, yes. I guess one way to present it is to say that the more
repeated measurments you make, the better you can approximate the
'true theoretical' value.

I have seen a preprint of Dr. Irizarry's work about the computation of
the expression values. He is surely the knowledgeable person to give
further details. 


>  For example, I've run a subset of data through the process, using chips hybridized with extract from untreated cells and cells exposed to various treatment.  I then analyzed the cells using 2 chips, 3 chips, 4 chips and 5 chips (ie. untreated + 1-4 treatments), and the results I obtained were different each time.  Below is the values generated for 10 genes in for the untreated condition only.
> 
>      gene 2 chips 3chips 4chips 5chips 
>       untreated 203307_at 8.47863 8.42976 8.213314 8.44459 
>      205793_x_at 9.063024 9.022368 8.84279 9.022368 
>      209168_at 7.840245 7.743475 7.584038 7.817369 
>      210260_s_at 10.42869 10.37537 10.02658 10.42127 
>      210373_at 8.373434 8.366322 8.076776 8.366322 
>      214692_s_at 8.729106 8.765005 8.457484 8.769177 
>      215294_s_at 6.884695 6.95363 6.691277 6.849249 
>      216017_s_at 8.90209 8.637539 8.472341 8.830136 
>      217431_x_at 7.288174 7.246028 6.960451 7.288174 
>      59437_at 7.960958 8.008521 7.547648 7.834895 
> 
> 
> I recognize that the values don't vary greatly, but I am concerned that having a treatment that causes profound effects to expression of certain genes may affect my results considerably  (with respect to not having had done that treatment).  
> 

Testing whether two groups of values vary can be considered different or
not is addressed by specific tests. The t-test has been for example widely used
with microarrays data. Alternatives can be found (the R-package 'permax'
offers such an alternative).

(note: In the case of Affymetrix arrays, for each gene several short probes
(oligonucleotides) are found on the array.
The values you present above are 'summary' values for all the probes values.
You may want to look at the probe level the reason the differences
(see the affy.pdf vignette). I have seen cases where an odd difference
in the expression level of experiments is caused by artefacts (some of
them being visible on the image of the chip (the package has function to
let you observe the physical location of probes related to a particular
gene (try 'demo(affy.tour)').



> 
> I hope my question is clear and would very much appreciate your comments on this.
> 


I hope this could help a bit....



Regards,



Laurent






> Regards
> Larry Heisler
> Dept of Laboratory Medicine and Pathobiology
> Program in Proteomics and Pathobiology
> University of Toronto
> Toronto, Ontario Canada
> l.heisler@rogers.com
> 
> 

-- 
--------------------------------------------------------------
Laurent Gautier			CBS, Building 208, DTU
PhD. Student			DK-2800 Lyngby,Denmark	
tel: +45 45 25 24 89		http://www.cbs.dtu.dk/laurent