[BioC] Comparison of diff. t-statistics, Limma and rowttests

Boel Brynedal Boel.Brynedal at ki.se
Fri Jul 25 09:33:26 CEST 2008

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-statistics
from rowttests ("Diff_tStatistics_Limma.jpg").

Am I doing something completely wrong? The assumption of equal variance
taken 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,

My code and sessionInfo:

# eset is a filtered, gcrma normalized ExpressionSet with ~10 000 probe
sets, 24 arrays.

ordinary.t <- fit3$coef / fit3$stdev.unscaled / fit3$sigma

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)


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

More information about the Bioconductor mailing list