[BioC] LIMMA help: how to choose correct weight for flagged spots??

daniela marconi daniela.marconi at libero.it
Tue Feb 28 17:38:46 CET 2006


Hi limma users!!!
Does anyone know why and how a different weight value for flagged spot vastly influence significance levels(for both p-value and log odds) and order in topTable() list?

I flagged spot with a self-made script for genePix(taking into acconunt some quality parameter like Signal to noise ratio and so on...)

After I've used different values as weight (in wtflags(): 1, 0, 0.1, 0.5, 0.01) obtaining really different results.
For example with 0 and 1 ...I obtained no significant adjusted p-values (P>=0.05)
For Weight=0.01 I obtained very signhificant P-value: for more or less 300 genes in the topTable list I have P-value in the range of  E-20 to E-10 
I'm wondering what is the correct choice for weights?


Thanks for any help and suggestion
Daniela

PS: (Below I attached my R script in wich the only change is in read.maimages(...,wt.fun=wtflag(???))

Marconi Daniela 
Phd Student 
Bologna University 
Physics Department
Viale Berti P. 6/2 
40137 Bologna

tel: +39 0512095136
e-mail: daniela.marconi at bo.infn.it


########Reading data


>library(limma)
>memory.limit(4000)
>Targets<-readTargets()
>RG<-read.maimages(Targets$file.name,source="genepix",wt.fun=wtflags(0.01))

>RG$printer<-getLayout(RG$genes)
>MA<-normalizeWithinArrays(RG,bc.method="minimum")
>MAlast<-normalizeBetweenArrays(MA,method="quantile")
>MAlast$targets<-Targets

>Targets

1    013.gpr nsM linf B
2    015.gpr  UM linf B
3    018.gpr  UM linf B
4    021.gpr  UM linf B
5    022.gpr  UM linf B
6    032.gpr  UM linf B
7    039.gpr  UM linf B
8    047.gpr  UM linf B
9    049.gpr nsM linf B
10   067.gpr  sM linf B
11   068.gpr nsM linf B
12   079.gpr  sM linf B
13   080.gpr nsM linf B
14   098.gpr  sM linf B
15   107.gpr  sM linf B
16   119.gpr  UM linf B
17   127.gpr  sM linf B
18   128.gpr  UM linf B
19   129.gpr nsM linf B
20   149.gpr  UM linf B
21   164.gpr nsM linf B
22   181.gpr  sM linf B
23   185.gpr  sM linf B
24   186.gpr  sM linf B
25   188.gpr nsM linf B
26   191.gpr  UM linf B
27   195.gpr  UM linf B
28   245.gpr  UM linf B
29   257.gpr  sM linf B
30   258.gpr nsM linf B
31   286.gpr nsM linf B
32   287.gpr  sM linf B
33   288.gpr nsM linf B
34   304.gpr nsM linf B
35   305.gpr  sM linf B
36   313.gpr nsM linf B
37   316.gpr  sM linf B
38   318.gpr  sM linf B
39   320.gpr  sM linf B
40   323.gpr nsM linf B
41   325.gpr  sM linf B
42   326.gpr nsM linf B
43   328.gpr  UM linf B
44   329.gpr  sM linf B
45   331.gpr  UM linf B
46   332.gpr  sM linf B
47   334.gpr  sM linf B
48   337.gpr  sM linf B
49   338.gpr  sM linf B
50   340.gpr  sM linf B
51   344.gpr  UM linf B
52   345.gpr  UM linf B
53   346.gpr nsM linf B
54   354.gpr  UM linf B
55   369.gpr nsM linf B
56   378.gpr nsM linf B
57   382.gpr nsM linf B


#######################
#######################               LIMMA
#######################


>group<-factor(c("M",rep("UM",7),"M","M","M","M","M",rep("M",2),"UM","M","UM","M","UM","M",
rep("M",3),"M",rep("UM",3),"M",rep("M",2),"M",rep("M",2),"M","M",rep("M",3),"M","M","M",
"UM","M","UM",rep("M",5),rep("UM",2),"M","UM",rep("M",3)),levels=c("UM","M"))


>design<-model.matrix(~0+group)
>colnames(design)<-c("M","UM")


>dupcor <- duplicateCorrelation(MAlast,design=design)

>fitCOR <- lmFit(MAlast,ndup=2,correlation=dupcor$consensus.correlation,design=design,weights=RG$weights)

>cont.matrix<-makeContrasts(UM.M=UM-M,levels=design)

>fit2COR<-contrasts.fit(fitCOR,cont.matrix)

>fit2COR<-eBayes(fit2COR)

>result<-topTable(fit2COR,n=300,sort.by="P",adjust.method="fdr")

>write.table(result,file="limmaUMvsMflag0.txt",quote = FALSE, row.names = FALSE,sep="\t")



More information about the Bioconductor mailing list