[BioC] Normalization

[w.huber at dkfz-heidelberg.de]

Hi -

> ... The problem is that microarrays really don't handle data very well
> where the gene is off in one channel and on in another.  By definition,
> off is zero, and we are obsessed by calculating ratios, where a zero
> value really screws things up :-(

Microarrays are pretty fine at measuring small intensities, as you say it
is just our taking ratios or logarithms that can potentially make it
problematic.

> Anyway, your options are:
> - adjust your whole distribution such that 95% of spots are > 0
> ... this is quite popular, though again, dubious in it's validity

Another option that is a little more valid is implemented in the vsn
package; see the accompanying Bioinformatics/ISMB 2002 paper for details.
Instead of shifting the distribution, we adjust the transformation (which
in some sense is equivalent), and we have some statistical arguments for
why this is a valid thing to do, and a proper model-based parameter
estimation procedure for actually doing it.

Best regards
  Wolfgang