Dear All, I'd like to continue on the topic I've posted a few daysa go ("unbalanced design") - but in a slightly different direction ... I'm trying to figure out a dose response from an experiment (affy) that was conducted in three independent studies (basically each study conducted the same dose treatment, using slightly different protocols). I'd like to find genes that are consitently modulated by dose across the three studies using a linear model: value ~ study + dose + study:dose The problem is that the studies have quite different intensity distributions of the PMs). I've some doubts that RMA or VSN are appropiate normalization methods for this problem. Forcing all distributions to be the same may buffer the very strong study effect but may also erase some of the tiny (but significant) differences due to dose. I think the linear model above should take care of study differences, so maybe a less stringent normalization would do it. I'd be interested what normalization people use for this type of analysis. kind regards, Arne

Dear Arne, > The problem is that the studies have quite different intensity > distributions of the PMs). I've some doubts that RMA or VSN are > appropiate normalization methods for this problem. Forcing all > distributions to be the same may buffer the very strong study > effect but may also erase some of the tiny (but significant) > differences due to dose. Actually, VSN does not enforce intensity distributions to be the same. The normalization in VSN amounts to an affine-linear mapping of the intensities of each chip, that is, for each chip only a multiplicative/scaling factor and an additive offset are estimated from the data. Best regards, Anja -- Dr. Anja von Heydebreck Global Technologies Bio- and Chemoinformatics Merck KGaA 64271 Darmstadt Germany Phone: +49-6151-723235, Fax: +49-6151-723329 Email: anja.von.heydebreck@merck.de http://www.merck.de