[BioC] loess normalisations within limma and marrayNorm disagree slightl y

Gordon Smyth smyth at wehi.edu.au
Wed Aug 27 02:04:44 MEST 2003


At 12:10 AM 27/08/2003, michael watson (IAH-C) wrote:
>... I can't put it much simpler than the subject... if I take the same set 
>of data and loess normalise it using both packages, accepting the 
>defaults, the values that come out differ by approximately +/- 0.02 each 
>time....

Which versions of the packages are you using? There are several different 
loess normalization methods, which one did you use? Are you using weights? 
It would be best to give the actual commands you used in the two packages.

>Is there a reason for that?  Perhaps some rounding up or something?

No there are other differences. In the latest versions of the packages, 
marrayNorm uses span=0.4 by default while limma uses span=0.3. This would 
easily produce the difference you see. (No doubt you have already checked 
the documentation for such an obvious cause as this.) Other very small 
differences arise because limma uses lowess when there aren't any weights 
and loess when there are, while marrayNorm uses loess throughout.

Gordon



More information about the Bioconductor mailing list