[BioC] affy express->"No convergence... too many outliers"

[Rafael A. Irizarry]

the li and wong algorithm, as described in their paper, doesnt always
converge. when you fit it to tens of thousands of probe sets you are bound
to get some cases where it doesnt coverege. i think you can change this
somewhat by playing with the maximum iteration and convergence criteria
arguments of li.wong 

hope this helps,
rafael
On Wed, 17 Jul 2002, Carleton Garrett wrote:

> I am working through the example in the documentation 'Text Description
> of affy' which creates an exprSet 'LiWong' using express() with the
> 'Dilution' dataset and summary.stat=li.wong).  I get the following
> messages:
> 
> Normalizing Data
> Background correcting
> Preparing Data
> Computing expression. This may take a while.
> Warning messages:
> 1: No convergence achieved, too many outliers in:
> fit.li.wong(t(data.matrix), remove.outliers, normal.array.quantile,
> 2: No convergence achieved in outlier loop
>  in: fit.li.wong(t(data.matrix), remove.outliers, normal.array.quantile,
> 
> 3: No convergence achieved, too many outliers in:
> fit.li.wong(t(data.matrix), remove.outliers, normal.array.quantile,
> 4: No convergence achieved, too many outliers in:
> fit.li.wong(t(data.matrix), remove.outliers, normal.array.quantile,
> 
> It doesn't seem to cause much of a problem with the Dilution dataset as
> I can get the 'correct' results when I test the resulting exprSet with
> the 'assess' function described in the document.
> 
> However, when I use a significantly bigger dataset (16 U133A chips) that
> I have collected into a plob class, I get 50+ similar warnings about
> outliers and the exprSet 'LiWong' is not created at the conclusion of
> express().
> 
> Any thoughts regarding the origin of these warnings or how to avoid
> them?
> 
> (I am working on a windows NT platform with R1.5.1 and all packages
> obtained after June 26, 256 RAM, >70Gb harddrive)
> 
> Thanks
> 
> Carleton Garrett
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@stat.math.ethz.ch
> http://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>