[BioC] (no subject)

[thama at imbb.forth.gr]

Dear List,

I am using limma to analyze a gene expression microarray experiment, similar to
the one described in the Monograph (Chapter 23.5: Technical Replication, page 404):
a knock-out versus wt (BIO effect) 2-color mouse experiment, using dye-swaps
(DYE effect) and two independent biological replicates (GROUP effect).

My problem is that I get a lot of differences of unexpectedly high significance,
even for expression changes as low as 1.4-fold (e.g. log2ratio=0.48, fdr=0.015,
see below).

My questions are:
 Is the design correct, as the result of the fitted model has zero degrees of
freedom?

 If so, can you comment the high significance of subtle expression changes?

 If not, 
 - should I choose which effect (DYE or GROUP) to include in the model, along
with the BIO one, based on their significance?
 - is the fitting itself considered a bug?


Details and code

  SlideNumber   Cy3	Cy5 
1        	wt1	ko1
2        	ko1	wt1 
3        	wt2	ko2  
4          	ko2	wt2 

# After normalizing the data using:
RG1none <-  backgroundCorrect(RG1, method="none")
MARG1none <-  normalizeWithinArrays(RG1none, method= "printtiploess")
MARG1none <- normalizeBetweenArrays(MARG1none, method= "Aquantile")

# using the design:
design <- cbind (ko1vswt1= c(1, -1, 0, 0), ko2vswt2= c(0, 0, 1, -1), dye= c(1,
1, 1, 1))

# subsetting the non-control reporters
MouseGenes <- MARG1none$genes$Status== "Mouse"

# fitting the models 
fit <- lmFit(MARG1none[MouseGenes, ], design)

cont.matrixBIO <- makeContrasts (KOvsWT= (ko1vswt1 + ko2vswt2)/2, levels= design)
cont.matrixGROUP <- makeContrasts (Group1vsGroup2= (ko1vswt1 - ko2vswt2),
levels= design)
cont.matrixDYE <- makeContrasts (DYE= dye, levels= design)

fitBIO <- contrasts.fit (fit, cont.matrixBIO)
fitGROUP <- contrasts.fit (fit, cont.matrixGROUP)
fitDYE <- contrasts.fit (fit, cont.matrixDYE)

fitBIO <- eBayes (fitBIO)
fitGROUP <- eBayes (fitGROUP)
fitDYE <- eBayes (fitDYE)

 >topTable(fitBIO, num=5000, adjust="fdr")
                   ^^^^^^^^
      Block Row Column            ID          Name Status     logFC  AveExpr
..
19095    29   6     14 MMAA300016782 MMAA300016782  Mouse  0.4823145 12.392166


               t      P.Value    adj.P.Val            B
..
19095   3.930997 2.383282e-03 1.523999e-02 -2.862820232




> R.version # Win XP SP2
               _                                          
platform       i386-pc-mingw32                            
..                                       
version.string R version 2.4.1 Patched (2007-01-10 r40440)

> .Platform$GUI # GNU Emacs 21.3.1,  ESS version 5.3.3
[1] "RTerm"

> package.version("limma")
[1] "2.9.11"


Thank you all and especially Dr Smyth for their wonderful work.
Best regards,
Thanasis

PS I am a biologist and this is my first post to the list.



Thanasis Margaritis
Institute of Molecular Biology and Biotechnology
FORTH, Vassilika Vouton
P.O.Box 1385
GR 711 10 Heraklion, Crete
GREECE
Tel: +30 2810 391927