Question

simultaneous testing in DESeq2

0

Entering edit mode

Koen Van den Berge ▴ 180

@koen-van-den-berge-6369

Last seen 6 months ago

Ghent University, Belgium

Suppose I am analysing a dataset with two timepoints, two conditions within every timepoint and three replicates for every time-treat combination. Furthermore suppose that the hypotheses of interest are (a) DE within timepoint1 (b) DE within timepoint2 (c) differential DE between timepoint1 and timepoint2, which is basically the difference in contrasts of contrast a and contrast b. The two factors could be coded as

treat=rep(c(0,1,0,1),each=3)

time=rep(0:1,each=6)

In edgeR, after fitting a model with ~treat+time+treat:time as the design, one can make a contrast matrix to simultaneously test all three hypotheses for every gene analogous to an ANOVA, with the null hypothesis: Is any of the three contrasts false?

How could this be done using DESeq2?

deseq2 • 958 views

ADD COMMENT • link updated 8.2 years ago by Michael Love 43k • written 8.2 years ago by Koen Van den Berge ▴ 180

score 1 · Answer 1 · 2016-09-19

1

Entering edit mode

Michael Love 43k

@mikelove

Last seen 1 day ago

United States

We don't have contrast matrices in DESeq2, but you can just build these three results tables just as easily:

results(dds, name="treat1_vs_0")
results(dds, contrast=list(c("treat1_vs_0","treat1.time1"))
results(dds, name="treat1.time1")

ADD COMMENT • link 8.2 years ago Michael Love 43k

0

Entering edit mode

Thank you, Michael. However your response tests every hypothesis separately.

I am rather interested in testing all three hypotheses simultaneously in an F-test like fashion, checking if any of the three are false with one omnibus test, rather than performing every test separately.

ADD REPLY • link 8.2 years ago Koen Van den Berge ▴ 180

0

Entering edit mode

We only have support for Wald tests and LRT. I don't see how you could formulate this as an LRT between a full and a reduced model.

ADD REPLY • link 8.2 years ago Michael Love 43k