[BioC] RMA results

[Rafael A. Irizarry]

On Wed, 22 Jan 2003, Duhaime Johanne wrote:

> Hello
> 
> We are using the last version (version 1.1) for windows. We just start to use the software.
> We had previously analysed our data with Affymetrix and GeneSpring. 
> 
> We used RMA method (rma()) to normalize and calculate the expression value of 5 version mgu74 (not V2) affy chips.
> We had some genes that had 5 to 10 fold change with Affy and GeneSpring. Now we got everything around 1 as  fold change with rma. 
> Is someone can explain me this result?

two things to keep in mind:
    
1) rma return log2 expression so make
sure to use 2^(exp1-exp2) to compute fold change.

2) rma sacrifices a bit in bias for big gains in precission. for example,
using affymetrix's spike in data and using the fold-change>2 criteria
(ignoring presence absent calls) to define differentially expressed genes,
  
    MAS 5.0 gives you, on average, 12.5 true positives (out of 14 truely
    differentially expressed genes) but 3110 false positives.
  
    RMA on the other hand gives you a bit smaller average # of true 
    positives, 11.6, but a much smaller average # of false positives: 18.

hope this helps,
rafael


> 
> Thank you in advance
> 
> Johanne Duhaime
> IRCM
> duhaimj@ircm.qc.ca
> tel: (514) 987-5556
> fax: (514) 987-5644
> www.ircm.qc.ca
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@stat.math.ethz.ch
> http://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>