[BioC] p.adjust versus decideTests time series

[Urska Cvek]

Hello,


I am using limma on a set of 4 time points (3, 6, 20 and 30) each of which has two replicates, for two tissues (wild type and modified), total of 16 arrays. Thus, I set the typical:

TS <- factor(TS, levels=c("w3.SV40", "w6.SV40", "w20.SV40", "w30.SV40", "w3.WT", "w6.WT", "w20.WT", "w30.WT"))
design <- model.matrix(~0+TS)
colnames(design) <- levels(TS)
fit <- lmFit(x, design)

I am interested in the probe sets that respond differently in SV40 versus the wild type. First, I would like to find the probe sets that are differentiated at either of the time points (at least one, identified as an OR relationship) and then the probe sets that are differentiated at each of the time points (intersection). I am considering my time point 3 as baseline:

cont.dif <- makeContrasts(
        Dif06w = (w6.SV40-w3.SV40)-(w6.WT-w3.WT),
        Dif20w = (w20.SV40-w3.SV40)-(w20.WT-w3.WT),
        Dif30w = (w30.SV40-w3.SV40)-(w30.WT-w3.WT), levels=design)
fit2.dif<- contrasts.fit(fit,cont.dif)
fit2.dif <- eBayes(fit2.dif)

I proceed with two things:

sel.dif<-p.adjust(fit2.dif$F.p.value, method="fdr")< 0.05
summary(sel.dif)
This returns 1944 probe sets that are "true" or differentially expressed.

results <- decideTests(fit2.dif, adjust.method="fdr", p.value=0.05)
vennDiagram(results)
This gives me the venn diagram with 1 probe setmatching for contrast 1, 214 for contrast 2, 3438 for contrast 3. Their intersection is a single probe set.


I am trying to determine why does the p.adjust return 1944 probe sets that have adjusted p value 
_________________________________________________________________

 the go.