Quoting Davide Valentini <davide.valentini at="" ki.se="">: > > > Hi to all, > > I've to deal with a dataset that has a huge amount of flagged as "bad" > spots. My data are from peptide microarrays and the proportion of flags > is around the 90% in each slide. Luckily I have a good set of samples > (35 cases and 35 controls), but I'm not an expert with this kind of > problem. Should I treat the flagged data as missing values and so try to > impute new values instead the flagged "bad" spots ? I know the KNN > imputation or the SVD imputation. > What is normally done with the spots flagged as "bad" (-100, following > GenePix criteria + additional criteria) in cDNA experiments ? > > Sorry if the question looks banal, but as I said I'm not an expert on > this field. Any help is useful, also links regarding flagged data > analysis... > > Thanks in advance, > > Davide Hi Davide, when looking at the actual images, are the high number of "bad" spots indicating artifacts on the slides (dirt, scratches, high and uneven background...) or just a reflection of your probes lighting up just a small proportion of the spots? Genepix flags as "Not Found" spots where there's no signal, but if the background is a little high and uneven, it sometimes will try hard to find a spot, and find a group of pixels to call a spot, which then often fails other quality criteria (shape-related) and teh spot is flagged as "Bad" rather than "Not Found". I think you should look at the scanned images to get a very good idea of what's going on. Regarding what to do with flagged spots... I tend to disregard GenePix flags. Ocassionally there'll be some dust or a bubble affecting the signal on a few spots in a given array. I just proceed, relying on the fact that I have replicates and that I hope I won't get two specs of dust on two arrays on teh exact same place. If a given array concerns me, and I need to use it in my analysis, I may then create some weights (like the Genepix flags) to exclude particular spots from a given slide. I do my analyses using the Limma package, which allows you to use weights at different levels. I find it useful to remove all spots that have negligible signal in both channels (2-colour data) on ALL arrays. In my latest experiments that amounts to up to 30% of the spots. I remove them completely. If you have a large % of spots in your arrays that won't light up with your samples, it may be a good idea to identify them and remove them from the analysis... If you end up having only a relatively small number of spots you will have to be extra careful about your normalisation procedure... but that's another matter. From what you describe, I would want to know first of all why I am getting so many "bad" spots: is there a problem with the hybridisation/slides, or do my samples only light up a small % of spots? Jose -- Dr. Jose I. de las Heras Email: J.delasHeras at ed.ac.uk The Wellcome Trust Centre for Cell Biology Phone: +44 (0)131 6513374 Institute for Cell & Molecular Biology Fax: +44 (0)131 6507360 Swann Building, Mayfield Road University of Edinburgh Edinburgh EH9 3JR UK -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

Sorry for the delay in my reply, But I haven't noticed you message before ! Thanks a lot for your suggestions. I submitted your indications to the bio lab, Me too, I was wondering why so many flags, we are discussing the situation with the group. Thanks a lot Davide J.delasHeras at ed.ac.uk wrote: > Quoting Davide Valentini <davide.valentini at="" ki.se="">: > > >> Hi to all, >> >> I've to deal with a dataset that has a huge amount of flagged as "bad" >> spots. My data are from peptide microarrays and the proportion of flags >> is around the 90% in each slide. Luckily I have a good set of samples >> (35 cases and 35 controls), but I'm not an expert with this kind of >> problem. Should I treat the flagged data as missing values and so try to >> impute new values instead the flagged "bad" spots ? I know the KNN >> imputation or the SVD imputation. >> What is normally done with the spots flagged as "bad" (-100, following >> GenePix criteria + additional criteria) in cDNA experiments ? >> >> Sorry if the question looks banal, but as I said I'm not an expert on >> this field. Any help is useful, also links regarding flagged data >> analysis... >> >> Thanks in advance, >> >> Davide >> > > > Hi Davide, > > when looking at the actual images, are the high number of "bad" spots > indicating artifacts on the slides (dirt, scratches, high and uneven > background...) or just a reflection of your probes lighting up just a > small proportion of the spots? > Genepix flags as "Not Found" spots where there's no signal, but if the > background is a little high and uneven, it sometimes will try hard to > find a spot, and find a group of pixels to call a spot, which then > often fails other quality criteria (shape-related) and teh spot is > flagged as "Bad" rather than "Not Found". I think you should look at > the scanned images to get a very good idea of what's going on. > > Regarding what to do with flagged spots... I tend to disregard GenePix > flags. Ocassionally there'll be some dust or a bubble affecting the > signal on a few spots in a given array. I just proceed, relying on the > fact that I have replicates and that I hope I won't get two specs of > dust on two arrays on teh exact same place. If a given array concerns > me, and I need to use it in my analysis, I may then create some > weights (like the Genepix flags) to exclude particular spots from a > given slide. I do my analyses using the Limma package, which allows > you to use weights at different levels. > I find it useful to remove all spots that have negligible signal in > both channels (2-colour data) on ALL arrays. In my latest experiments > that amounts to up to 30% of the spots. I remove them completely. If > you have a large % of spots in your arrays that won't light up with > your samples, it may be a good idea to identify them and remove them > from the analysis... If you end up having only a relatively small > number of spots you will have to be extra careful about your > normalisation procedure... but that's another matter. > > From what you describe, I would want to know first of all why I am > getting so many "bad" spots: is there a problem with the > hybridisation/slides, or do my samples only light up a small % of spots? > > Jose > > -- Davide Valentini PhD - Biostatistician Department of Medical Epidemiology and Biostatistic Karolinska Institute Box 281 SE-171 77 Stockholm SWEDEN Tel: +46-8-524 82294 Fax: +46-8-31 4975