[BioC] negative values in raw data

[Wolfgang Huber]

Hi Guillaume,

some image processing / background correction methods produce negative 
estimates. In many cases, the "vsn" method from the package of the same 
name can deal with this. It uses the "glog" transformation, and glog(x) 
is similar to log(x+x0) with appropriate choice of x0. Please read the 
vignette carefully and use the diagnostic plot "meanSdPlot" to see 
whether it makes sense on your data.

While true gene abundances are obviously always >=0, it can make sense 
to have negative estimates if you have an error model that includes a 
Normally (or similar) distributed additive noise term. Because if Y is 
the observation, x is the true value, and eps the noise,

     Y = x + eps

will be negative half of the times if x=0.

Bw
  Wolfgang

  Deplaine wrote:
> Hi,
> 
>    I downloaded data on the GEO web site and now, I try to analyze them 
> with R 2.0.1 for windows XP but in this dataset, there is a lot of 
> negative values. So when I normalyze with log2 transformation, R 
> produced NaNs. I understand why but after, when I want to calculate 
> p-value with the parametric t test, R can't. I tried to write 0 where 
> there is a negative value :  R calculate p-value and q-value but all of 
> q-value are at 1.
> 
>    Is there someone to help me.
> Thanks a lot
> 
-------------------------------------
Wolfgang Huber
European Bioinformatics Institute
European Molecular Biology Laboratory
Cambridge CB10 1SD
England
Phone: +44 1223 494642
Fax:   +44 1223 494486
Http:  www.ebi.ac.uk/huber