Dear all, I am analyzing data from 44k agilent microarrays in limma. I have several issues where I'd like to hear some opinions from others. First, I have put the weights of the controlspots to 0 after reading in RG, in the command to normalize, is that ok? Second, using agilent FE flags, not recommended I read? Should anything be flagged out beforehand, like signals <0? And what about the saturated ones? Third, I have been experimenting with several background correction methods. It turns out that 'normexp' gives good results for most, but not all arrays. Two of them look really weird, with the MA cloud only starting after 5.. I came to the conclusion that the method Subtract results in the best looking MA plots for all of my arrays. The 'half' and 'minimum' method do not give such good results, with ugly stripes. But now (ofcourse after within (loess) and between (aquantile) normalization), I am not able to do 'single channel analysis' because the intraspot correlation cannot process missing or infinite M or A values.. what to do? Fourth, when I was experimenting with single channel analysis I was able to do a groupwise analysis, but the paired-samples analysis did not work because of 'no residual degrees of freedom' error. We have hybridized for each control and each experimental subject 1 baseline sample (-, cy3) and 1 challenged (+, cy5) sample. So we want to compare the +/- ratio for each experimental subject with a specifically matched control, is that possible? Hope to read advise from you soon, Simone de Jong

Quoting Simone de Jong <dejong_simone at="" hotmail.com="">: > Third, I have been experimenting with several background correction > methods. It turns out that 'normexp' gives good results for most, but > not all arrays. Two of them look really weird, with the MA cloud only > starting after 5.. I came to the conclusion that the method Subtract > results in the best looking MA plots for all of my arrays. The 'half' > and 'minimum' method do not give such good results, with ugly stripes. Have you also tried 'none', no background correction at all? I have recently found that some of my slides give much better results if I do not correct for background. The slides are very clean with very low background and good signals. It turns out that the slides that benefited the most from background correction wre all scanned with an ArraywoRx scanner, and the other ones were scanned with an Axon 4200AL, so it can be largely a scanner-dependent effect (I haven't pursued this because the ArraywoRx one is in another building so I just use now the one closest to me, two floors down). Maybe it does nothing to you, but if your slides are all similar (hybs, scanner, etc) maybe 'none' will be the best option. Give it a try if you haven't yet. 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