Hi all: I found a strange situation. Before I did normalization, the correlation within slide is pretty high, 0.5009. After I did normalization, the correlation goes down to 0.1672. I am pretty sure spacing between duplicates is correct. Does it mean the normalization method is not correct? I am using printtiploess normalization. > cor <- duplicateCorrelation(RG, design, ndups=6, spacing=1152) > cor$consensus.correlation [1] 0.5009 > cor <- duplicateCorrelation(MA.norm, design, ndups=6, spacing=1152) > cor$consensus.correlation [1] 0.1672 I highly appreciate any comments. Hua Weng -----Original Message----- From: bioconductor-bounces@stat.math.ethz.ch [mailto:bioconductor-bounces at stat.math.ethz.ch] On Behalf Of Gordon Smyth Sent: Thursday, November 17, 2005 10:55 PM To: Devin Scannell Cc: BioC Mailing List Subject: [BioC] duplicateCorrelation Dear Devin, There are a couple of problems. Firstly, you've told us that your replicates are 112 spots apart, but you haven't told limma this. So the software is assuming that the replicates are side-by-side, which is the default. You need instead: > cor <- duplicateCorrelation(MA, design, ndups=3, spacing=112) Secondly, two arrays is pretty minimal to estimate duplicate correlations. The help page for duplicateCorrelation says: For this function to return statistically useful results, there must be at least two more arrays than the number of coefficients to be estimated, i.e., two more than the column rank of 'design'. Hence you need at least 3 arrays to have confidence in your results whereas you have only two. If you want to check that duplicateCorrelation() is getting the right input, the best way is to check that your replicates really are at the spacing you think they are. Your data files (ScanArray?) almost certainly contain a gene ID column. Let's assume this column is called "ID". Use > RG <- read.maimages(..., annotation="ID") Then > unwrapdups(MA$genes$ID, ndups=3, spacing=112) is a matrix which should have three identical columns. Does it? Best wishes Gordon >[BioC] duplicateCorrelation >Devin Scannell scannedr at tcd.ie >Fri Nov 18 02:03:07 CET 2005 > >Hi, > >this is not a very interesting question but it has given me enough >trouble to get me to mail the list so I hope somebody has time to >reply. > >I have several two-colour arrays to analyze. Each probe is present >three times on each chip and they are spaced 112 spots apart (not my >decision). The consensus correlation returned by duplicateCorrelation >is typically around zero which is surprising since the spots are close >together and the data looks good in MA plots (even before >normalization). A histogram of the individual correlations >(cor$all.correlations from duplicateCorrelation) supports the >conclusion that the within-chip replicates are poorly correlated. > >I am concerned that the numbers that are being handed to >duplicateCorrelation are incorrect somehow but I am not sure what I am >doing wrong (code below). I have looked at the code for >duplicateCorrelation and cannot follow it so I was wondering if anyone >can suggest a way to verify the correlations it is calculating. Ideally >I would like to be able to select a specific gene, calculate the >correlation between replicates myself and verify that this is the same >as I obtain from duplicateCorrelation. > >Thanks in advance, >Devin >library(limma) > >targets <- readTargets() > >targets > SlideNumber Name FileName Cy3 Cy5 >13 13 60H_9:12 13.csv WT1 60H1 >17 17 60H_12:9 17.csv 60H1 WT1 > >flag.check <- function(x) as.numeric(x$Flags >= 3) >RG <- read.maimages(targets$FileName, sep=",", columns=list(Rf="Ch1 >Median",Gf="Ch2 Median",Rb="Ch1 B Median",Gb="Ch2 B Median"), >wt.fun=flag.check) > >RG$genes <- readGAL() >RG$printer <- getLayout(RG$genes) > >RG.bgc <- backgroundCorrect(RG, method="normexp", offset=50) >MA <- normalizeWithinArrays(RG.bgc, method="loess") > >design <- cbind(c(1,-1)) >cor <- duplicateCorrelation(MA, design, ndups=3) _______________________________________________ Bioconductor mailing list Bioconductor at stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/bioconductor

Dear Hua, At 09:57 AM 3/12/2005, Hua Weng wrote: >Hi all: > >I found a strange situation. Before I did normalization, the correlation >within slide is pretty high, 0.5009. After I did normalization, the >correlation goes down to 0.1672. Doesn't seem at all strange to me. That's what is supposed to happen. You can't really expect anything sensible from standard deviations or correlations on un-normalized data. > I am pretty sure spacing between duplicates >is correct. Does it mean the normalization method is not correct? No, quite the opposite. BTW, your question would be more meaningful if you tried to explain why you find the behaviour surprising, and what you expected instead. Best wishes Gordon > I am using >printtiploess normalization. > > > cor <- duplicateCorrelation(RG, design, ndups=6, spacing=1152) > > cor$consensus.correlation >[1] 0.5009 > > cor <- duplicateCorrelation(MA.norm, design, ndups=6, spacing=1152) > > cor$consensus.correlation >[1] 0.1672 > >I highly appreciate any comments. > >Hua Weng