[BioC] marray, weights and normalizations..

Gordon Smyth smyth at wehi.edu.au
Thu Apr 21 09:54:14 CEST 2005

At 05:28 PM 21/04/2005, Henning Redestig wrote:
>Another issue has occured though which I have seen on several datasets now 
>related to using zero weights. Distributionally I get a whole lot more 
>outliers leading to M values ranging between e.g. -200, 200 an effect I 
>cant see when using weights of say, 0.1 instead (for all negatively 
>GenePix flagged genes). Is this to be expected or am I doing something wrong?

Yes it is to be expected. The meaning of zero weights is that there is no 
penalty for the loess line not fitting these spots, and hence it is to be 
expected that it tends to fit these spots poorly. Hence large normalized 

Of course, if you really believe that a spot should get zero weight, you 
shouldn't care what M-value it gets, because it will play no role in the 
analysis. Because of this, the plotMA() function in limma hides spots with 
zero weights as a default.

Btw, I am very much opposed to the idea that a spot should be considered 
poor quality merely because it is faint, e.g., gets a GenePix -50 flag. If 
a gene is not expressed in a particular sample, a faint spot is exactly 
what you want to observe.


>Thanks for the reply!

More information about the Bioconductor mailing list