decideTests and topTable
1
0
Entering edit mode
@humberto-ortiz-zuazaga-1209
Last seen 10.2 years ago
I've got some Agilent microarray data for 3 different groups, and am trying to use limma to select differentially expressed genes. Here's the steps I've taken so far: > allone <-function (qta) + { + 1 + } > targets <- readTargets("new-cort-targets.txt") > RG <- read.maimages(targets$FileName,source="agilent",wt.fun=allone) Read 1/Cort(N-C)2.txt Read 2/cort(N-E)2.txt Read 2/cort(N-E)3.txt Read 5/dye swap/cort(C-N).txt Read 5/dye swap/Cort(E-N).txt > types <- readSpotTypes("spottype.txt") > status <- controlStatus(types, RG) Matching patterns for: ControlType Found 22393 other Found 913 Positive Found 162 Negative Found 21318 gene Setting attributes: values Color ID Name > RG$genes$Status <- status > RG <- backgroundCorrect(RG, method="none") > weights <- modifyWeights(RG$weights, status, + values=c("Positive","Negative"),multipliers=0) > MA <- normalizeWithinArrays(RG,method="loess",weights=weights) > MA.b <- normalizeBetweenArrays(MA,method="scale") > design <- modelMatrix(targets, ref="Naive") Found unique target names: Control Enriched Naive > fit <- lmFit(MA.b,design,weights=weights) > fit.b <- eBayes(fit) > table <- topTable(fit.b,coef=1,number=100) > write.table(table,file="table.txt", sep="\t", col.names = NA) > calls.strict <- decideTests(fit.b,adjust.method="fdr") > write.fit(fit.b,results=calls.strict,file="fit.txt",digits=3) My question is, when I look at the top table, my best candidate is "" "Row" "Col" "ProbeUID" "ControlType" "ProbeName" "GeneName" "Description" "Status" "M" "A" "t" "P.Value" "B" "10984" 52 106 10181 0 "A_51_P443387" "AJ276707" "Mus musculus partial mRNA for WTAP protein" "gene" -0.9176259 9.403058 -11.262342 0.2490771 -2.218647 Which has an adjusted p value of 0.2490771 The fit object also has a p-value column, and it is adjusted in write.fit, but the corresponding line from the fit is: A Control Enriched Control Enriched Control Enriched Control Enriched Row Col ProbeUID ControlType ProbeName GeneName Description Status 9.40 -0.918 -0.267 -11.26 -4.02 0.00001 0.00533 -1 0 52 106 10181 0 A_51_P443387 AJ276707 Mus musculus partial mRNA for WTAP protein gene The p value for contrast 1 is 0.00001. Why are the p values so different? Can I say this gene is or is not differentially expressed? Note that the decideTests result for contrast 1 is -1, so I understand that decideTests thinks it is differentially expressed. Looking at the topTable output, however, makes it unlikely to be differentially expressed. -- Humberto Ortiz Zuazaga Programmer-Archaeologist High Performance Computing facility University of Puerto Rico http://www.hpcf.upr.edu/~humberto/
Microarray Mus musculus limma Microarray Mus musculus limma • 1.6k views
ADD COMMENT
0
Entering edit mode
@gordon-smyth
Last seen 2 hours ago
WEHI, Melbourne, Australia
Are you clear on what an adjusted p-value is? When you used topTable() you used the default adjustment method which is "holm", a very conservative method. When you used decideTests() you used adjustment method "fdr", a much less conservative method. When you looked directly at the fitted model object, the p-value is not adjusted at all. It is not therefore not surprising that the p-values are different in the three cases. Gordon >Date: Tue, 19 Apr 2005 10:53:31 -0400 >From: Humberto Ortiz Zuazaga <humberto@hpcf.upr.edu> >Subject: [BioC] decideTests and topTable >To: Bioconductor@stat.math.ethz.ch > >I've got some Agilent microarray data for 3 different groups, and am >trying to >use limma to select differentially expressed genes. > >Here's the steps I've taken so far: > > > allone <-function (qta) >+ { >+ 1 >+ } > > targets <- readTargets("new-cort-targets.txt") > > RG <- read.maimages(targets$FileName,source="agilent",wt.fun=allone) >Read 1/Cort(N-C)2.txt >Read 2/cort(N-E)2.txt >Read 2/cort(N-E)3.txt >Read 5/dye swap/cort(C-N).txt >Read 5/dye swap/Cort(E-N).txt > > types <- readSpotTypes("spottype.txt") > > status <- controlStatus(types, RG) >Matching patterns for: ControlType >Found 22393 other >Found 913 Positive >Found 162 Negative >Found 21318 gene >Setting attributes: values Color ID Name > > RG$genes$Status <- status > > RG <- backgroundCorrect(RG, method="none") > > weights <- modifyWeights(RG$weights, status, >+ values=c("Positive","Negative"),multipliers=0) > > MA <- normalizeWithinArrays(RG,method="loess",weights=weights) > > MA.b <- normalizeBetweenArrays(MA,method="scale") > > design <- modelMatrix(targets, ref="Naive") >Found unique target names: > Control Enriched Naive > > fit <- lmFit(MA.b,design,weights=weights) > > fit.b <- eBayes(fit) > > table <- topTable(fit.b,coef=1,number=100) > > write.table(table,file="table.txt", sep="\t", col.names = NA) > > > calls.strict <- decideTests(fit.b,adjust.method="fdr") > > write.fit(fit.b,results=calls.strict,file="fit.txt",digits=3) > >My question is, when I look at the top table, my best candidate is > >"" "Row" "Col" "ProbeUID" "ControlType" "ProbeName" >"GeneName" "Description" >"Status" "M" "A" "t" "P.Value" "B" >"10984" >52 106 10181 0 "A_51_P443387" "AJ276707" "Mus >musculus partial mRNA >for WTAP >protein" "gene" -0.9176259 9.403058 -11.262342 >0.2490771 -2.218647 > >Which has an adjusted p value of 0.2490771 > >The fit object also has a p-value column, and it is adjusted in write.fit, >but >the corresponding line from the fit is: > >A Control Enriched Control Enriched Control >Enriched Control Enriched Row Col >ProbeUID ControlType ProbeName GeneName >Description Status > 9.40 -0.918 -0.267 -11.26 -4.02 0.00001 0.00533 > -1 0 52 106 10181 0 >A_51_P443387 AJ276707 Mus musculus partial mRNA for WTAP >protein gene > >The p value for contrast 1 is 0.00001. > >Why are the p values so different? > >Can I say this gene is or is not differentially expressed? Note that the >decideTests result for contrast 1 is -1, so I understand that decideTests >thinks it is differentially expressed. Looking at the topTable output, >however, makes it unlikely to be differentially expressed. > >-- >Humberto Ortiz Zuazaga >Programmer-Archaeologist >High Performance Computing facility >University of Puerto Rico >http://www.hpcf.upr.edu/~humberto/
ADD COMMENT

Login before adding your answer.

Traffic: 950 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6