glm edgeR issue

[giuliano stirparo]

Hi everybody
I just performed a time series analysis on edgeR and I have a question.
This is my data frame for the condition and time:

sample condition time
1 sample1 Placebo 1h
2 sample2 Placebo 1h
3 sample3 Placebo 2h
4 sample4 Placebo 2h
3 sample5 Placebo 4h
4 sample6 Placebo 4h
5 sample7 Drug 1h
6 sample8 Drug 1h
7 sample9 Drug 2h
8 sample10 Drug 2h
9 sample11 Drug 4h
10 sample12 Drug 4h

and my script to test interaction

design<-model.matrix(~condition*time,data=meta)
y<-DGEList(counts=countTable)
y<-calcNormFactors(y,method="TMM")
y<-estimateGLMCommonDisp(y,design)
y<-estimateGLMTrendedDisp(y,design)
y<-estimateGLMTagwiseDisp(y,design)
fit<-glmFit(y,design)
colnames(design)
"(intercept)" "conditionDrug" "time2h" "time4h" "conditionDrug:time2h" "conditionDrug:time4h"
lrt<-glmLRT(fit,coef=6)

In this case with coef=6 I am able to detect the genes that respond differently to the drug, relative to placebo at 4h.
But same gene that are modulated are not expressed either in placebo 4h and Drug 4h!!!
Example(matrix of countTable)

ID Placebo-1h Placebo-1h Placebo-2h Placebo-2h Placebo-4h Placebo-4h Drug-1h Drug-1h Drug-2h Drug-2h Drug-4h Drug-4h
ENSMUSG000000xxxx 0 5 4 0 0 0 0 3 0 1 0 0

And here the different comparison

ID logFC logCPM LR PValue
"ENSMUSG000000xxxx" -4.36 0.14 6.48 0.01 Placebo4-h
"ENSMUSG000000xxxx" -2.19 0.14 1.21 0.27 Drug-4h
"ENSMUSG000000xxxx" 2.16 0.14 100.14 1.41e-23 Contrast(Placebo-4h/Drug-4h)

As you can see in the contrast Placebo-4h and Drug-4h I have a logFC of 2.16(that is the difference between Placebo-1h/Placebo-4h and Drug-1h/Drug-4h) and a PValue of 1.41e-23!!But I have no reads on this gene!
How is possible?
Could you help me to understand if I made same mistake in the analysis?

Thank you

Giuliano

[Gordon Smyth]

No this isn't possible, so you have presumably made a mistake.

The edgeR User's Guide asks you to send questions to the Bioconductor mailing list. If you write to that list, someone will help diagnose your problem.