Hi, I have a time course experiment in which three genptypes' RNAs (wt, mu1, mu2) are extracted at 5 time points. I used common reference design. For time point 0, I have six arrays, cy3 cy5 array1 ref wt_Notreatment array2 wt_Notreatment ref array3 ref mu1_Notreatment array4 mu1_Notreatment ref array5 ref mu2_Notreatment array6 mu2_Notreatment ref For time point from 1 to 4, each time point has 12 arrays. six arrays are like above, samples are without treatment. And the other six arrays are for samples with treatment. I did the following steps, design<-modelMatrix(targets, ref="ref") design<-cbind(Dye=1,design) pair<-rep(1:27,each=2) corfit<-duplicateCorrelation(MA,design,block=pair) Warning message: NaNs produced in: atanh(rho) traceback() No traceback available corfit$consensus [1] -1 what does Warning message mean here? and why corfit$consensus is -1? I used R version 2.0.0 and limma version 1.8.12, then I upgraded limma to version 1.9.0, and I got same Warning message from each version of limma. Any help will be greatly appreciated. Thanks!! Ren [[alternative HTML version deleted]]

> Date: Thu, 12 May 2005 14:39:48 -0500 > From: "Na, Ren" <na@uthscsa.edu> > Subject: [BioC] warning message from duplicateCorrelation in limma > To: <bioconductor@stat.math.ethz.ch> > > Hi, > I have a time course experiment in which three genptypes' RNAs (wt, mu1, mu2) are extracted at 5 > time points. > I used common reference design. > > For time point 0, I have six arrays, > cy3 cy5 > array1 ref wt_Notreatment > array2 wt_Notreatment ref > array3 ref mu1_Notreatment > array4 mu1_Notreatment ref > array5 ref mu2_Notreatment > array6 mu2_Notreatment ref > > For time point from 1 to 4, each time point has 12 arrays. six arrays are like above, samples are > without treatment. And the other six arrays are for samples with treatment. > > I did the following steps, > > design<-modelMatrix(targets, ref="ref") > design<-cbind(Dye=1,design) > pair<-rep(1:27,each=2) > corfit<-duplicateCorrelation(MA,design,block=pair) > Warning message: > NaNs produced in: atanh(rho) > traceback() > No traceback available > corfit$consensus > [1] -1 > > what does Warning message mean here? and why corfit$consensus is -1? > I used R version 2.0.0 and limma version 1.8.12, then I upgraded limma to version 1.9.0, and I got > same Warning message from each version of limma. > Any help will be greatly appreciated. > Thanks!! > > Ren >From your description of your experiment you seem to have confounded blocks entirely with treatment effects, and it just isn't possible to sensibly estimate a within-block correlation in this situation. I'm actually a bit disappointed that duplicateCorrelation() gives you have value here for $consensus: I would prefer it gave you NA, which it did when I tried to reproduce your result. Why are you setting 'block=pair'? This seems to imply that you have no biological replication for any of your treatments, which would be a very serious design flaw for your experiment. Gordon

________________________________ From: Gordon K Smyth [mailto:smyth@wehi.EDU.AU] Sent: Sat 5/14/2005 3:50 AM To: Na, Ren Cc: bioconductor@stat.math.ethz.ch Subject: [BioC] warning message from duplicateCorrelation in limma > Date: Thu, 12 May 2005 14:39:48 -0500 > From: "Na, Ren" <na@uthscsa.edu> > Subject: [BioC] warning message from duplicateCorrelation in limma > To: <bioconductor@stat.math.ethz.ch> > > Hi, > I have a time course experiment in which three genptypes' RNAs (wt, mu1, mu2) are extracted at 5 > time points. > I used common reference design. > > For time point 0, I have six arrays, > cy3 cy5 > array1 ref wt_Notreatment > array2 wt_Notreatment ref > array3 ref mu1_Notreatment > array4 mu1_Notreatment ref > array5 ref mu2_Notreatment > array6 mu2_Notreatment ref > > For time point from 1 to 4, each time point has 12 arrays. six arrays are like above, samples are > without treatment. And the other six arrays are for samples with treatment. > > I did the following steps, > > design<-modelMatrix(targets, ref="ref") > design<-cbind(Dye=1,design) > pair<-rep(1:27,each=2) > corfit<-duplicateCorrelation(MA,design,block=pair) > Warning message: > NaNs produced in: atanh(rho) > traceback() > No traceback available > corfit$consensus > [1] -1 > > what does Warning message mean here? and why corfit$consensus is -1? > I used R version 2.0.0 and limma version 1.8.12, then I upgraded limma to version 1.9.0, and I got > same Warning message from each version of limma. > Any help will be greatly appreciated. > Thanks!! > > Ren >From your description of your experiment you seem to have confounded blocks entirely with treatment effects, and it just isn't possible to sensibly estimate a within-block correlation in this situation. I'm actually a bit disappointed that duplicateCorrelation() gives you have value here for $consensus: I would prefer it gave you NA, which it did when I tried to reproduce your result. Dr. Gordon Smith, I thought "corfit<-duplicateCorrelation(MA,design,block=pair)" calculates the correlation between every slide and its dye swap ( a pair of technical replicates). I have 27 pairs of technical replicates here. Maybe I misunderstand the section " technical replicate" of User's guide. Please enlight me if I did anything wrong. Then I can use lmFit and makeContrast to find out differential expression between any comparison I want to make. Why are you setting 'block=pair'? This seems to imply that you have no biological replication for any of your treatments, which would be a very serious design flaw for your experiment. Yes, You are right. We had no biological replication in this experiment. This is a initial experiment. All RNA samples are extracted from cell culure dishs. The purpose of the experiment is try to find out differential expression between time points and that between genotypes, and locate which time point our interested genes changes. Then we focuse on that period time to do futher time point experiment. Gordon [[alternative HTML version deleted]]

Dr. Smyth, If I want to use the argument block, the blocks in the pair should correspond to biological replicates, but mine is not. So I can't use block. Is that what you mean. Yes, you are right. we don't have biological replicates. All samples are extracted from cell culture dishes. Thanks! Ren -----Original Message----- From: Gordon K Smyth [mailto:smyth@wehi.EDU.AU] Sent: Sat 5/14/2005 3:50 AM To: Na, Ren Cc: bioconductor@stat.math.ethz.ch Subject: [BioC] warning message from duplicateCorrelation in limma > Date: Thu, 12 May 2005 14:39:48 -0500 > From: "Na, Ren" <na@uthscsa.edu> > Subject: [BioC] warning message from duplicateCorrelation in limma > To: <bioconductor@stat.math.ethz.ch> > > Hi, > I have a time course experiment in which three genptypes' RNAs (wt, mu1, mu2) are extracted at 5 > time points. > I used common reference design. > > For time point 0, I have six arrays, > cy3 cy5 > array1 ref wt_Notreatment > array2 wt_Notreatment ref > array3 ref mu1_Notreatment > array4 mu1_Notreatment ref > array5 ref mu2_Notreatment > array6 mu2_Notreatment ref > > For time point from 1 to 4, each time point has 12 arrays. six arrays are like above, samples are > without treatment. And the other six arrays are for samples with treatment. > > I did the following steps, > > design<-modelMatrix(targets, ref="ref") > design<-cbind(Dye=1,design) > pair<-rep(1:27,each=2) > corfit<-duplicateCorrelation(MA,design,block=pair) > Warning message: > NaNs produced in: atanh(rho) > traceback() > No traceback available > corfit$consensus > [1] -1 > > what does Warning message mean here? and why corfit$consensus is -1? > I used R version 2.0.0 and limma version 1.8.12, then I upgraded limma to version 1.9.0, and I got > same Warning message from each version of limma. > Any help will be greatly appreciated. > Thanks!! > > Ren >From your description of your experiment you seem to have confounded blocks entirely with treatment effects, and it just isn't possible to sensibly estimate a within-block correlation in this situation. I'm actually a bit disappointed that duplicateCorrelation() gives you have value here for $consensus: I would prefer it gave you NA, which it did when I tried to reproduce your result. Why are you setting 'block=pair'? This seems to imply that you have no biological replication for any of your treatments, which would be a very serious design flaw for your experiment. Gordon [[alternative HTML version deleted]]

Is does not make sense to fit the same factor as both random (block) and fixed (in the design matrix), which is what you are doing. It doesn't make sense from a scientific, statistical or computing point of view. You have two choices: (i) simply average the dye swaps and compute fold changes between treatments or (ii) use lmFit() with an apriori fixed correlation, e.g., around 0.3, which would be similar to (i) but would down-weight probes with disagreement between dye swaps. In the latter case, you should also include a dye-effect term. Gordon At 07:58 AM 18/05/2005, Na, Ren wrote: >Dr. Smyth, > >If I want to use the argument block, the blocks in the pair should >correspond to biological replicates, >but mine is not. So I can't use block. Is that what you mean. > >Yes, you are right. we don't have biological replicates. All samples are >extracted from cell culture dishes. > >Thanks! > >Ren