Joyce: The choice of normalization is heavily depended on the biological assumptions. Usually, I will do location normalization first, followed by scale normalization. To clarify the assumptions behind, let us consider the following extreme examples: 1) Suppose there is a drug affecting the OVERALL transcription activities. As a consequence, ALL genes are expressed half as the level of normal control. If we do location normalization, we will not see the effect of this drug! 2) Suppose there is a drug that can modulate the expression of x number of the genes on the chip. As a consequence, the magnitude of these x genes is twice as normal control (no directional changes, just an amplitude change). If we do a scale normalization, this effect might be erased and unseen. 3) Suppose we are using a focused array with only 100 apoptosis- related genes. If drug X induces apoptosis, and thus most of the genes on this chip are upregulated; we can find ourselves in a more difficult position. Solutions include special controls for normalization. Simon ================================================= Simon M. Lin, M.D. Manager, Duke Bioinformatics Shared Resource Assistant Research Professor, Biostatistics and Bioinformatics Box 3958, Duke University Medical Center Durham, NC 27710 Ph: (919) 681-9646 FAX: (919) 681-8028 Lin00025 (at) mc.duke.edu http://dbsr.duke.edu ================================================= > Message: 1 > Date: Fri, 30 Jan 2004 09:02:04 -0600 > From: Joyce Gu <jwgu@bcm.tmc.edu> > Subject: [BioC] Normalization with cDNA array > To: bioconductor@stat.math.ethz.ch > Message-ID: <401A72AC@webmail.bcm.tmc.edu> > Content-Type: text/plain; charset="ISO-8859-1" > > Hello, > There are two ways to normalize the data, one is location normalization, the > other is scale normalization in marrayClasses. I am wondering what kind of > situation should we use location normalization,and when should we use scale > normalization? > Does anyone has any suggession? > > Thanks >