Thanks for your answer! Another question more the community in general: Do you think we can ask Affymetrix to send us another chip when we do have a considerable artifact which is only visible with the help of the false color image? Here we already did that when the chip was having a visible scratch on the scan picture (= the regular one in B&W). David On Nov 17, 2005, at 17:46, Ben Bolstad wrote: > Hi David, > > To actually decide whether or not to remove a chip from a dataset the > recommended method is to use either or both of the RLE and NUSE > boxplots > and look for arrays with boxes that are centered higher or more spread > than the others. > > The chip pseudo-images give you the ability to potentially diagnose the > cause of the lesser quality data. Generally speaking, smaller artifacts > are not problematic (and probably aren't even reflected in the NUSE or > RLE) although they can sometimes be interesting as can be seen by some > of the images in the gallery. This is because on most modern chips the > probes making up each probeset are distributed all across the array, > which means that there is typically only one probe in any probeset > affected. Thus provided you are using a robust expression measure eg > RMA > or the expression values from fitPLM itself you won't have any problem. > > Larger artifacts are more problematic (and usually you'll find > correspond with NUSE and RLE) because it typically means there are some > probesets with a large number of probes being affected. If it is really > bad you probably don't have much choice but to remove the whole chip > from subsequent analysis. > > If you feel you are somewhere in between then your best option is > probably to do some down-weighting of lesser quality data. I believe if > you do your analysis using limma and supply the PLMset from fitPLM it > will do something like this, though you might have other ideas about > how > to do the weighting. I think this would be a better approach then just > completely masking the area. > > > Thanks, > > Ben > > > On Thu, 2005-11-17 at 10:32 +0100, David Ruau wrote: >> Dear Dr. Bolstad, >> >> Thanks for those really interesting gallery, it make you feel less >> alone when you have a look at your own chips after. >> I would like to know if a tool exist to remove the badly hybridized >> zone of the chip from the analysis? >> I mean just keeping the good zone of the chip or do not include the >> bad >> zone into the rest of the analysis. >> >> I motivated by this problem because my boss always gnash one's teeth >> when I say that this chip is bad and we should not use it. >> >> David >> >> On Nov 10, 2005, at 1:03, Ben Bolstad wrote: >> >>> The gallery of quality images from fitPLM has moved to >>> >>> http://plmimagegallery.bmbolstad.com/ >>> >>> As always, contributions are welcome. >>> >>> _______________________________________________ >>> Bioconductor mailing list >>> Bioconductor at stat.math.ethz.ch >>> https://stat.ethz.ch/mailman/listinfo/bioconductor >>> >> >> _______________________________________________ >> Bioconductor mailing list >> Bioconductor at stat.math.ethz.ch >> https://stat.ethz.ch/mailman/listinfo/bioconductor > >