Actually, I did not express myself well. I was not talking about error correlation. People usually look at the correlation in M between arrays. Only the differentially expressing genes should be correlated - the remainder of M-values are just noise and so should be uncorrelated across arrays. For this reason the size of the correlation should be small. Since M represents A - B on one array and B - A on the other (with treatments A and B), the correlation should be negative. Great suggestions for reading. --Naomi At 08:54 AM 3/13/2010, James W. MacDonald wrote: >Hi Ana, > >You need to re-read what Naomi said. A correlation between dye swaps >would be expected. What she was warning about was a correlation >between treatments. > >As for a good book, have you read the BioC monograph? > >http://bioconductor.org/pub/docs/mogr/ > >The case studies might be of interest as well > >http://bioconductor.org/pub/biocases/ > >Best, > >Jim > > > >Ana Staninska wrote: >>Dear Naomi, Thank you very very much for your very helpful answers. >>Could you maybe tell me what did you mean that if the dye swap >>correlation between two treatments is -0.2 I am in trouble. What is >>considered to be a good dye swap correlation (I calculate it using >>duplicateCorrelation function in limma). Also what is considered as >>a good correlation between duplicate spots (after normalization) ? >>I know that the easiest way out is to ask a statistician to do the >>analysis, but I would like to learn it myself to do it (I am a >>mathematician, so I think I should be able to learn it). Could you >>maybe point out a literature that I could read and learn a proper >>way of dealing with any kind of microarrays. >>Thank you very much one more time, Best, Ana >> >>>Date: Fri, 12 Mar 2010 16:29:03 -0500 >>>To: staninska at hotmail.com; naomi at stat.psu.edu; >>>bioconductor at stat.math.ethz.ch >>>From: naomi at stat.psu.edu >>>Subject: RE: [BioC] (no subject) >>> >>>Dear Ana, >>>I actually meant that you should average dye swaps, not spots, >>>although either is OK as long as you use corfit for the other. >>> >>>If there are no technical replicates for some biological reps, the >>>analysis is much more complicated. This really requires a >>>statistical consultant and someone who will do some detailed >>>preliminary analyses. >>> >>>Naomi >>> >>>p.s. I hope that the correlation of -0.2 for the dye swaps is for >>>R-G. If it is for treatment A - treatment B, you have a problem. >>> >>>At 03:08 PM 3/12/2010, Ana Staninska wrote: >>>>Dear Naomi, >>>> >>>>Thank you very much for your answer. I just have few follow up question. >>>> >>>>How big should be the correlation on my duplicate spots in order >>>>to "safetly" average them? >>>>Before the normalization, the correlation on my duplicate spots >>>>is around 0.7-0.8, but after normalization >>>>it is only around 0.4-0.6. Which I think it is not the best. >>>>Probably I should mention that the correlation of dye swapped >>>>arrays is around -0.2. >>>> >>>>Also, for some of the experiments, we had to remove certain >>>>arrays, and therefore not all of my biological replicates are dye swapped. >>>>In that case I think I should use the contrast matrix to average >>>>of the treated vs non-treated comparisons. >>>>Isn't then better to use the corfit$consensus on my duplicate spots? >>>> >>>>Thank you very much in advance, >>>> >>>>All the best, >>>>Ana >>>> >>>> >>>> >>>> >>>> >>>>>Date: Fri, 12 Mar 2010 12:28:06 -0500 >>>>>To: staninska at hotmail.com; bioconductor at stat.math.ethz.ch >>>>>From: naomi at stat.psu.edu >>>>>Subject: Re: [BioC] (no subject) >>>>> >>>>>The estimated error variance used for the test denominator will be an >>>>>average of technical and biological replication, and therefore not >>>>>really appropriate for your analysis. However, you could average the >>>>>2 technical replicates prior to running limma which would give you >>>>>the right error structure. >>>>> >>>>>--Naomi >>>>> >>>>>At 12:04 PM 3/12/2010, Ana Staninska wrote: >>>>> >>>>>>Dear Bioconductor, >>>>>>I have a simple experiment that I have to analyze in order to find >>>>>>differentially expressed genes. I have 10 biological replicates, and >>>>>>each biological replicate has two technical replicates which appear >>>>>>as dye swapped. So in total I have 20 arrays. Each of the probes are >>>>>>spotted twice on the array (on the left and on the right hand side). >>>>>>I use limma to do my analysis. I know at the moment it is not >>>>>>possible to treat duplicate spots, technical replicates and >>>>>>biological replicates, but I though if I use the >>>>>>duplicateCorrelation function on my duplicate spots, and then to use >>>>>>a contrast matrix to average of all of the Treated vs Non- treated >>>>>>biological samples, I could address all 3 replications. Am I correct? >>>>>> >>>>>> >>>>>>I am sending a copy of my code, if someone could look at it at tell >>>>>>me whether I made somewhere a mistake. >>>>>>Thank you very much in advance, >>>>>>Sincerely Ana Staninska >>>>>> >>>>>> >>>>>>library(limma)> library(statmod)> library(marray)> >>>>>>library(convert)> library(hexbin)> library(gridBase)> >>>>>>library(RColorBrewer)> > targets <- >>>>>>readTargets("Lysi_270705.txt")> > ### Only manually removed ot >>>>>>absent spots are given 0 weight ###> RGa <- read.maimages(targets, >>>>>>source="genepix", wt.fun=wtflags(weight=0, >>>>>>cutoff=-75), other.columns=c("F635 SD","B635 SD","F532 SD","B532 >>>>>>SD","B532 Mean","B635 Mean","F Pixels","B Pixels"))Read >>>>>>LYSI270705_1_200905.gpr Read LYSI270705_1dw_200905.gpr Read >>>>>>LYSI270705_2_200905.gpr Read LYSI270705_2dw_200905.gpr Read >>>>>>LYSI270705_3_121005.gpr Read LYSI270705_3dw_121005.gpr Read >>>>>>LYSI270705_4_121005.gpr Read LYSI270705_4dw_121005.gpr Read >>>>>>LYSI270705_5_121005.gpr Read LYSI270705_5dw__121005.gpr Read >>>>>>LYSI270705_6_121005.gpr Read LYSI270705_6dw__121005.gpr Read >>>>>>LYSI270705_7_151001.gpr Read LYSI270705_7dw_151005.gpr Read >>>>>>LYSI270705_8_151005.gpr Read LYSI270705_8dw_151005.gpr Read >>>>>>LYSI270705_9_151005.gpr Read LYSI270705_9dw_151005.gpr Read LYSI270705! >>>>>>_10_151005.gpr Read LYSI270705_10dw_151005.gpr > for(i in >>>>>>1:nrow(RGa)){+ for(j in >>>>>>1:ncol(RGa)){+ if(RGa$Rb[i,j]+RGa$R[i,j]+ RGa$G[i,j]+ >>>>>>RGa$Gb[i,j] ==0)+ RGa$weights[i,j]<-0+ }+ }> > >>>>>>####################################################> ### >>>>>>Background Correction = Normexp + offset 25 ####> >>>>>>####################################################> > RG >>>>>><-backgroundCorrect(RGa, method="normexp", , normexp.method="mle", >>>>>>offset=25)Green channelCorrected array 1 Corrected array 2 >>>>>>Corrected array 3 Corrected array 4 Corrected array 5 Corrected >>>>>>array 6 Corrected array 7 Corrected array 8 Corrected array 9 >>>>>>Corrected array 10 Corrected array 11 Corrected array 12 Corrected >>>>>>array 13 Corrected array 14 Corrected array 15 Corrected array 16 >>>>>>Corrected array 17 Corrected array 18 Corrected array 19 Corrected >>>>>>array 20 Red channelCorrected array 1 Corrected array 2 Corrected >>>>>>array 3 Corrected array 4 Corrected array 5 Corrected array 6 >>>>>>Corrected array 7 Corrected array 8 Corrected array ! >>>>>>9 Corrected array 10 Corrected array 11 Corrected array 12 Corrected a >>>>>>rray 13 Corrected array 14 Corrected array 15 Corrected array 16 >>>>>>Corrected array 17 Corrected array 18 Corrected array 19 Corrected >>>>>>array 20 > ####################################################> >>>>>>##### normalize Within arrays #########> >>>>>>####################################################> > MA >>>>>><-normalizeWithinArrays(RG, method="loess")> > >>>>>>####################################################> ###### >>>>>>Contrast Matrix ############> >>>>>>####################################################> > >>>>>>design<-cbind( + MU1vsWT1=c( >>>>>>1,-1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),+ MU2vsWT2=c(0,0, >>>>>>1,-1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),+ MU3vsWT3=c(0,0,0,0, >>>>>>1,-1,0,0,0,0,0,0,0,0,0,0,0,0,0,0),+ MU4vsWT4=c(0,0,0,0,0,0, >>>>>>1,-1,0,0,0,0,0,0,0,0,0,0,0,0),+ MU5vsWT5=c(0,0,0,0,0,0,0,0, >>>>>>1,-1,0,0,0,0,0,0,0,0,0,0),+ MU6vsWT6=c(0,0,0,0,0,0,0,0,0,0, >>>>>>1,-1,0,0,0,0,0,0,0,0), >>>>>>+ MU7vsWT7=c(0,0,0,0,0,0,0,0,0,0! >>>>>>,0,0, >>>>>>1,-1,0,0,0,0,0,0),+ MU8vsWT8=c(0,0,0,0,0,0,0,0,0,0,0,0,0,0, >>>>>>1,-1,0,0,0,0),+ MU9vsWT9=c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, >>>>>>1,-1,0,0),+ MU10vsWT10=c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, >>>>>>1,-1))> > cont.matrix <- >>>>>> >>>>>>####################################################> >>>>>>### Duplicate Correlations on duplicate spots ####> >>>>>>####################################################> > >>>>>>corfit<-duplicateCorrelation(MA, ndups=2, spacing=192)> > >>>>>>####################################################> ##### Linear >>>>>>Fit Model and Contrasts fit #######> >>>>>>####################################################> > >>>>>>fit<-lmFit(MA, design, ndups=2, spacing=192, >>>>>>cor=corfit$consensus)> > fit<-contrasts.fit(fit, cont.matrix)> > >>>>>>####################################################> >>>>>>######### eBayes Statistics ###############> #################! >>>>>>###################################> > fit<-eBayes(fit)> > ########### >>>>>>###################################################> ### Writing >>>>>>the Results ######> >>>>>>##############################################################> >>>>>>TTnew<-topTable(fit,coef=1, number=100, adjust="BH") >>>>>> >>>>>> >>>>>>Ana StaninskaHelmholtz-Zentrum MuenchenDepartment of Scientific >>>>>>ComputingNeuherberg, Deutschland+49 (0) 89 3187 2656 >>>>>> >>>>>>[[alternative HTML version deleted]] >>>>>> >>>>>>_______________________________________________ >>>>>>Bioconductor mailing list >>>>>>Bioconductor at stat.math.ethz.ch >>>>>>https://stat.ethz.ch/mailman/listinfo/bioconductor >>>>>>Search the archives: >>>>>>http://news.gmane.org/gmane.science.biology.informatics.conducto r >>>>>Naomi S. Altman 814-865-3791 (voice) >>>>>Associate Professor >>>>>Dept. of Statistics 814-863-7114 (fax) >>>>>Penn State University 814-865-1348 (Statistics) >>>>>University Park, PA 16802-2111 >>>Naomi S. Altman 814-865-3791 (voice) >>>Associate Professor >>>Dept. of Statistics 814-863-7114 (fax) >>>Penn State University 814-865-1348 (Statistics) >>>University Park, PA 16802-2111 >> >> [[alternative HTML version deleted]] >>_______________________________________________ >>Bioconductor mailing list >>Bioconductor at stat.math.ethz.ch >>https://stat.ethz.ch/mailman/listinfo/bioconductor >>Search the archives: >>http://news.gmane.org/gmane.science.biology.informatics.conductor >********************************************************** >Electronic Mail is not secure, may not be read every day, and should >not be used for urgent or sensitive issues >_______________________________________________ >Bioconductor mailing list >Bioconductor at stat.math.ethz.ch >https://stat.ethz.ch/mailman/listinfo/bioconductor >Search the archives: >http://news.gmane.org/gmane.science.biology.informatics.conductor Naomi S. Altman 814-865-3791 (voice) Associate Professor Dept. of Statistics 814-863-7114 (fax) Penn State University 814-865-1348 (Statistics) University Park, PA 16802-2111