Hello lists, I am using Bioconductor/Limma package to analyze microRNA arrays. I have 7 replicate two-color arrays, 4 probe replicates each array. Follow Limma procedures, I had a list from topTable with LogFC(1) and other columns. In addition, I produced LogFC for each individual arrays from MAlist, the average LogFC(2) were then taken from 7 arrays to compare LogFC(1). The LogFC(2) is similar to LogFC(1), for example 3.12 vs 3.2, but they're not exactly the same. My question is that why aren't these two columns of value the same? Did I calculate it mistakenly or something behind this during the Limma procedures? Could anyone kindly explain it? Thanks. Kevin [[alternative HTML version deleted]]

Dear Kevin, Can I gently say, both to you and to other listers, that is not helpful to ask Bioconductor developers to justify why they do whatever they currently do without explaining why you expected the behaviour to be otherwise. The direct answer to your question is that different spots on any microarray normalize to different values because it is highly improbable they would give the same value. Measurement error and technical varialibity ensures that all values will be slightly different, even if they are probing the same gene or probe-sequence. Were you expecting normalization to summarize the replicate probe values? That is not the purpose of normalization. Normalization is intended to preserve information, not hide it. Summarization would be done by a subsequent analysis step, such as lmFit(). Any normalization method which set all replicate values equal would prevent downstream software from knowing whether or not the replicates were truly consistent. That would be misleading verging on dishonest. Best wishes Gordon > Date: Thu, 23 Oct 2008 13:25:46 -0700 (PDT) > From: Hai Lin <kevinvol2002 at="" yahoo.com=""> > Subject: [BioC] LogFC in Limma > To: bioconductor at stat.math.ethz.ch > Message-ID: <696005.24645.qm at web32404.mail.mud.yahoo.com> > Content-Type: text/plain > > Hello lists, > > I am using Bioconductor/Limma package to analyze microRNA arrays. > I have 7 replicate two-color arrays, 4 probe replicates each array. > Follow Limma procedures, I had a list from topTable with LogFC(1) and > other columns. In addition, I produced LogFC for each individual arrays > from MAlist, the average LogFC(2) were then taken from 7 arrays to > compare LogFC(1). > > The LogFC(2) is similar to LogFC(1), for example 3.12 vs 3.2, but > they're not exactly the same. > > My question is that why aren't these two columns of value the same? Did > I calculate it mistakenly or something behind this during the Limma > procedures? > > Could anyone kindly explain it? > > Thanks. > > Kevin

Dear Kevin, After sending my previous reply, I re-read your email and realized your question is probably about a coefficient from lmFit() rather than a question about your probe replicates. It is impossible to give you a definite answer because you haven't told us what what your design is or what logFC(1) is measuring, what model you're fitting, whether you used contrasts.fit and so on. Ordinarily, people will give enough code to make these things clear. Since you appear to have at least three columns in your design matrix, it seems rather unlikely that logFC(1) could be a simple average of all the arrays. Also, do you have quality weights in your data or blocks in your fit? In these cases, limma will compute weighted averages, which are not the same as ordinary averages. Best wishes Gordon > Date: Thu, 23 Oct 2008 13:25:46 -0700 (PDT) > From: Hai Lin <kevinvol2002 at="" yahoo.com=""> > Subject: [BioC] LogFC in Limma > To: bioconductor at stat.math.ethz.ch > Message-ID: <696005.24645.qm at web32404.mail.mud.yahoo.com> > Content-Type: text/plain > > Hello lists, > > I am using Bioconductor/Limma package to analyze microRNA arrays. > I have 7 replicate two-color arrays, 4 probe replicates each array. > Follow Limma procedures, I had a list from topTable with LogFC(1) and > other columns. In addition, I produced LogFC for each individual arrays > from MAlist, the average LogFC(2) were then taken from 7 arrays to > compare LogFC(1). > > The LogFC(2) is similar to LogFC(1), for example 3.12 vs 3.2, but > they're not exactly the same. > > My question is that why aren't these two columns of value the same? Did > I calculate it mistakenly or something behind this during the Limma > procedures? > > Could anyone kindly explain it? > > Thanks. > > Kevin