[BioC] Link to picture: Comparison of diff. t-statistics, Limma and rowttests

[Boel Brynedal]

Dear List,

Thank you (Wolfgang and Paolo) for telling me the attachment did not get
through. This is a link to the picture:
http://picasaweb.google.se/Boelbubblan/Statistics/photo

Cheers,
Boel

--~*~**~***~*~***~**~*~--
Boel Brynedal, MSc, PhD student
Karolinska  Institutet
Department of Clinical neuroscience


----- Original Message -----
From: Boel Brynedal <Boel.Brynedal at ki.se>
Date: Friday, July 25, 2008 9:33 am
Subject: Comparison of diff. t-statistics, Limma and rowttests
To: bioconductor at stat.math.ethz.ch

> Dear List,
> 
> I have affy hgu133plus2 arrays from individuals with disease, in two
> different stages of the disease. I've earlier used rowttests and FDR
> correction. Now I was playing around with limma to see what I could do
> (added different covariates etc) but also investigated the most simple
> setting, comparing the two different stages directly using Limma. The
> first thing that struck me was that limma "finds" only half the amount
> of significantly diff expressed genes. So I started to look at the
> t-statistics from limma. Then I stumbled across this: when I do a
> qq-plot of the ordinary t-statistics they are far from normally
> distributed, and actually totally strange. See attached plot comparing
> the ordinary t, the moderate t (both from Limma) as well as t-
> statisticsfrom rowttests ("Diff_tStatistics_Limma.jpg").
> 
> Am I doing something completely wrong? The assumption of equal 
> variancetaken using ordinary t could not create this, could it? 
> Please help me
> figure out what's wrong here, I'm hoping I've done some stupid 
> mistake.What else could explain this? Thank you.
> 
> Best wishes,
> Boel
> 
> My code and sessionInfo:
> 
> # eset is a filtered, gcrma normalized ExpressionSet with ~10 000 
> probesets, 24 arrays.
> library(limma)
> library(Biobase)
> library(genefilter)
> specific<-factor(c(rep("stageA",10),rep("stageB",14)),
> levels=c("stageB","stageA"))
> design<-model.matrix(~specific)
> fit<-lmFit(eset,design)
> Fit<-eBayes(fit)
> 
> ordinary.t <- fit3$coef / fit3$stdev.unscaled / fit3$sigma
> moderate.t<-Fit$t[,2]
> rowttests.t<-rowttests(eset,fac=specific)
> 
> par(mfrow=c(1,3))
> qqnorm(ordinary.t,main="fit ordinary.t")
> qqnorm(moderate.t, main=" Fit moderate.t")
> qqnorm(rowttests.t[,1], main= "rowttests.t")
> dev2bitmap("Diff_tStatistics_Limma.jpg",type="jpeg", height = 5, 
> width =
> 15, res = 75)
> 
> > sessionInfo()
> R version 2.7.1 (2008-06-23)
> x86_64-unknown-linux-gnu
> 
> locale:
> ...
> 
> attached base packages:
> [1] splines   tools     stats     graphics  grDevices utils     
> datasets[8] methods   base
> 
> other attached packages:
> [1] genefilter_1.20.0 survival_2.34-1   Biobase_2.0.1     limma_2.14.5
> 
> loaded via a namespace (and not attached):
> [1] annotate_1.18.0     AnnotationDbi_1.2.2 DBI_0.2-4
> [4] RSQLite_0.6-9
> 
> --~*~**~***~*~***~**~*~--
> Boel Brynedal, MSc, PhD student
> Karolinska  Institutet
> Department of Clinical neuroscience
> 
>