[BioC] Question regarding bg correction/normalization/summarization

[Wolfgang Huber]

References: <5B9643DA-D133-11D8-9F16-000A95B445D6 at ucla.edu>


Hi Tova,

with 'calibration' I refer to the transformation

     ny[k,i] <-  (y[k,i] - a[s[k], i])  /  b[s[k], i]

where y[k,i] is the raw intensity of the k-th probe on the i-th array,
s[k] is the "stratum" (e.g. print-tip group) of the k-th probe (in the
most simple case, there is just one stratum: s[k]==1 for all k),
a[s, i] is an offset term for the s-th stratum on the i-th array,
b[s, i] is a scaling factor for the s-th stratum on the i-th array, and
ny[k,i] is the calibrated intensity.

Don't be afraid of the formal notation, it's really very simple. The
parameters a and b are estimated from the data. It is in that sense that
vsn does background correction.

People use also the words "background correction" and "normalization" in
this context, but my impression is that the exact definition of what
they mean is a little different for each author.

 From my own prejudice, I would not recommend to use the combination
   bgcorrect.method="rma", normalize.method="vsn"
in expresso, since this would be doing the same thing twice (and
probably get confused) but I'd interested to hear about other opinions.

Also, note that the current example for the use of vsn with affy data
ignores the MMs, which is clearly suboptimal compared to methods that do
use the MMs in a sensible manner.

(Note: in addition, vsn will apply a "log-like" tranformation on the
matrix ny to ensure approximate independence of variance of the mean.
This transformation is called the "glog" and is like the logarithm (base
e) for high intensities and like a straight line for low intensities.)

Best wishes
  Wolfgang

-- 
-------------------------------------
Wolfgang Huber
Division of Molecular Genome Analysis
German Cancer Research Center
Heidelberg, Germany
Phone: +49 6221 424709
Fax:   +49 6221 42524709
Http:  www.dkfz.de/abt0840/whuber
-------------------------------------


Tova Fuller wrote:
> Hello!
> 
> I was wondering about the terminology used to describe VSN in its 
> vignette.  The vignette claims VSN does calibration, and mentions 
> something about comparison against background - so, in addition to 
> normalization, does VSN do background correction?  The example given for 
> use with the affy package has bg correction turned off, but I've notice 
> several posters have used rma for background correction with vsn in 
> expresso.
> 
> Also, I was curious as to opinions regarding which possible combinations 
> in expresso or threestep (affyPLM), in addition to DChip & PLIER are 
> most common for doing bg correction, normalization & summarization.  Or 
> which combos are the best.  Any feedback would be wonderful.
> 
> Thank you, and I apologize for these probably trivial questions - I am 
> but a lowly graduate student.
> 
> Thanks again,
> 
> Tova Fuller



-- 
-------------------------------------
Wolfgang Huber
Division of Molecular Genome Analysis
German Cancer Research Center
Heidelberg, Germany
Phone: +49 6221 424709
Fax:   +49 6221 42524709
Http:  www.dkfz.de/abt0840/whuber