Hello, I have a curious problem involving Limma. I have an ExpressionSet object (called Seminoma) that contains the results of 18 samples (12 tumours and 6 normals). The only strange thing I have done is to join Affymetrix U133A and B results (renaming the probes so that there is no overlap). > design [1] 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 I run the following to get the adjusted p-values for differential expression. > fit <- lmFit(Seminoma,design) > fit <- eBayes(fit) > tempo <- topTable(fit,n=nrow(exprs(Seminoma)),adjust="BH") > tail(tempo) ID logFC AveExpr t P.Value adj.P.Val B 36857 236888_at 4.050545 5.896910 2.848388 0.007679017 0.007679872 -2.489166 39997 240028_at 3.590545 5.153103 2.795185 0.008761283 0.008762063 -2.602027 5638 206112_at 4.702957 7.202596 2.749623 0.009800000 0.009800654 -2.697717 37129 237160_at 3.356252 4.824457 2.721845 0.010488579 0.010489046 -2.755610 9742 210262_at 4.657111 7.232671 2.690246 0.011326586 0.011326838 -2.821047 23669 223673_at 3.837049 5.837806 2.648655 0.012524609 0.012524609 -2.906478 As you can see, all the probes appear to be significantly differentially expressed. I am sure this should not be the case, especially after examining a number of different probes. For example, on probe "117_at" if I run a t-test() it produces a p-value of 0.1034 (no adjustment) whereas limma suggests it is 8.4e-07 (or 2.45e-06). My only thought is that something must be happening in the eBayes step. Can anyone help me on what I might be doing wrong? Many thanks Dan PS > sessionInfo() R version 2.4.1 (2006-12-18) x86_64-unknown-linux-gnu locale: LC_CTYPE=en_GB;LC_NUMERIC=C;LC_TIME=en_GB;LC_COLLATE=en_GB;LC_MONETARY =en_GB;LC_MESSAGES=en_GB;LC_PAPER=en_GB;LC_NAME=C;LC_ADDRESS=C;LC_TELE PHONE=C;LC_MEASUREMENT=en_GB;LC_IDENTIFICATION=C attached base packages: [1] "tools" "stats" "graphics" "grDevices" "utils" "datasets" [7] "methods" "base" other attached packages: Biobase limma "1.12.2" "2.9.13" -- ************************************************************** Daniel Brewer, Ph.D. Institute of Cancer Research Email: daniel.brewer at icr.ac.uk ************************************************************** The Institute of Cancer Research: Royal Cancer Hospital, a charitable Company Limited by Guarantee, Registered in England under Company No. 534147 with its Registered Office at 123 Old Brompton Road, London SW7 3RP. This e-mail message is confidential and for use by the addre...{{dropped}}

On Thursday 12 April 2007 06:24, Daniel Brewer wrote: > Hello, > > I have a curious problem involving Limma. I have an ExpressionSet > object (called Seminoma) that contains the results of 18 samples (12 > tumours and 6 normals). The only strange thing I have done is to join > Affymetrix U133A and B results (renaming the probes so that there is no > overlap). > > > design > > [1] 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 > > I run the following to get the adjusted p-values for differential > expression. > > > fit <- lmFit(Seminoma,design) > > fit <- eBayes(fit) > > tempo <- topTable(fit,n=nrow(exprs(Seminoma)),adjust="BH") > > tail(tempo) > > ID logFC AveExpr t P.Value adj.P.Val > B 36857 236888_at 4.050545 5.896910 2.848388 0.007679017 0.007679872 > -2.489166 39997 240028_at 3.590545 5.153103 2.795185 0.008761283 > 0.008762063 -2.602027 5638 206112_at 4.702957 7.202596 2.749623 > 0.009800000 0.009800654 -2.697717 37129 237160_at 3.356252 4.824457 > 2.721845 0.010488579 0.010489046 -2.755610 9742 210262_at 4.657111 > 7.232671 2.690246 0.011326586 0.011326838 -2.821047 23669 223673_at > 3.837049 5.837806 2.648655 0.012524609 0.012524609 -2.906478 > > As you can see, all the probes appear to be significantly differentially > expressed. I am sure this should not be the case, especially after > examining a number of different probes. For example, on probe "117_at" > if I run a t-test() it produces a p-value of 0.1034 (no adjustment) > whereas limma suggests it is 8.4e-07 (or 2.45e-06). > > My only thought is that something must be happening in the eBayes step. > Can anyone help me on what I might be doing wrong? Your design vector will test that the probes are significantly different from 0. You probably want to include an intercept term or (I find it more natural for two groups) define the two groups explicitly and then use a contrast matrix to get the difference between the two. In either case, your design matrix needs at least two columns. Sean