Hi all, I am very doubtful about the "control" method of normalizeWithinArray function in limma package. Suppose that I have an MAList object ma contains a two-color array: w = rep(1, dim(ma$M)[1]) w[ abs(ma$M) > 3] = 0 What is the difference between normalizeWithinArray(ma, method="control", layout=layout, controlspots= as.logical(abs(1-w)) ) and normalizeWithinArray(ma, method="loess", weights=w) ? In practice, the first one with method="control" is very very slow, while the second one is very quick. The motivation of my trying these methods is that I want to do loess normalization with only those probes that were not heavily differential-expressed. Could you help me out? Thanks in advance. Leon

Dear Leon, method="control" does what you want exactly whereas weights=w is an approximation. The exact method takes longer. See http://genomebiology.com/2007/8/1/R2 Weights=w is an approximation because the loess curve setup uses all the points to define the neighbourhood regions, even those for which w=0, and so the loess curve definition is not the same as would you like to define for the control spots alone. method="control" is slow because it is necessary to store the loess fit and then to compute predicted values for the non-control spots. Best wishes Gordon > Date: Sun, 04 Jan 2009 19:11:17 +0800 > From: Leon Yee <yee.leon at="" gmail.com=""> > Subject: [BioC] normalizeWithArray with method of "control" in limma > package > To: bioconductor at stat.math.ethz.ch > Message-ID: <49609955.9080502 at gmail.com> > Content-Type: text/plain; charset=ISO-8859-1; format=flowed > > Hi all, > > I am very doubtful about the "control" method of > normalizeWithinArray function in limma package. > Suppose that I have an MAList object ma contains a two-color array: > > w = rep(1, dim(ma$M)[1]) > w[ abs(ma$M) > 3] = 0 > > What is the difference between > normalizeWithinArray(ma, method="control", layout=layout, > controlspots= as.logical(abs(1-w)) ) > > and > > normalizeWithinArray(ma, method="loess", weights=w) > > ? > > In practice, the first one with method="control" is very very slow, > while the second one is very quick. > > The motivation of my trying these methods is that I want to do loess > normalization with only those probes that were not heavily > differential-expressed. > > Could you help me out? Thanks in advance. > > > Leon