[BioC] VSN: narrowing down probe sets for parameter estimation

[Wolfgang Huber]

Hi Stefan,

0) vsn already has an algorithm that attempts to narrow down the probes 
that are used. This is the so-called "robustification" of the ML 
estimator by Least Trimmed Sum of Squared minimisation. But of course 
this is automatic, and sometimes not perfect, and if you have an 
external way of identifiýing non-changing probes, that can be very useful.

1) A few hundred should be OK in practice. What is more important than 
their number is that they about equally cover the whole dynamic range!

2)
x = an ExpressionSet
fit = vsn2(x[yourSelectedProbes, ])
nx = predict(fit, newdata=exprs(x))

see also the man page

    method?predict("vsn")

(Please use latest release version.)


Hope this helps
  Wolfgang


> Dear all,
> 
> I am working on an Affymetrix time series data set with high percentages 
> (30-40%) and mostly downregulated differentials.
> 
> In a previous discussion regarding the question of a suitable normalization 
> strategy for such data sets Wolfgang Huber highly recommended to "narrow 
> down the probes from which you fit the parameters from all genes (incl. the 
> differential ones) to a subset which are enriched for non-changing."
> 
> In this context I have two questions:
> 
> 1) What is the minimum number of genes/probes that should be used for VSN 
> parameter estimation? I could extract a list of some hundred 'stable' or 
> 'low variability' genes from previous microarray studies. Would this number 
> be sufficient or do I need bigger probe subsets (thousands of probes, 1/2 
> of all probes, etc.)?
> 
> 2) Is there a straight foward way to implement this into standard R 
> packages offerring VSN? In other words, if I perform a VSN parameter 
> estimation on my gene/probe subset, how (in R terms) would I subsequently 
> apply this to the whole dataset?(My apologies if this is trivial, my 
> programming skills are still rather a disgrace :) )
> 
> Any comment on these questions would be highly appreciated.
> 
> Kind regards,
> 
> Stefan
>