Re: Interaction categorical/continuous variable DESeq2

Cher Hugo

sorry if I missed something, but why not fit the model with interactions an=
d test for the coefficient of the =91X' main effect?
(see arguments =91name=92, =91contrast=92 of the =91results=92 function).

What you propose below seems not wrong, but perhaps unnecessarily complicat=
ed.

	Best wishes
	Wolfgang

Il giorno 15 Sep 2014, alle ore 15:41, Hugo Varet <[email protected]> h=
a scritto:

> Dear list, dear Mike Love,
> =

> I am using DESeq2 to model counts from an unusual type of experiment and =
I have a question about the strategy I employed. The experiment consisted i=
n sequencing 33 samples for which we have the following information:
> - group (16 samples from group A and 17 from group B)
> - a continuous variable X almost uniform (variable of interest)
> =

> I have to add the group to the design formula because I know it has a str=
ong effect on the counts. Then, as my goal is to detect genes which vary wi=
th the continous variable X in the same way within both groups A and B, I w=
ant to exclude genes for which there is an interaction between group and X.=
 The design is thus ~ group + X + group:X and I used the following lines to=
 test the interaction:
> =

> dds <- DESeqDataSetFromMatrix(countData=3Dcounts, colData=3Dtarget, desig=
n =3D ~ group + X + group:X)
> dds <- estimateSizeFactors(dds)
> dds <- estimateDispersions(dds)
> dds <- nbinomWaldTest(dds)
> res <- results(dds, name=3D"groupB.X")
> sum(res$padj<=3D0.05, na.rm=3DTRUE)
> hist(res$padj)
> =

> As I found no significant interaction (the minimum adjusted p-value is ab=
out 0.6), I decided to remove the interaction term from the design and to u=
se ~ group + X. I can then test for the coefficients of X.
> =

> If I do not detect any significant interaction, I think it is due to a la=
ck of power. So, can I use the additive model ~ group + X even if it will n=
ot be correct for genes which actually have an interaction?
> =

> Many thanks in advance,
> =

> Hugo
> =

> PS: I am using R 3.1.1 and DESeq2 1.4.5
> =

> _______________________________________________
> Bioconductor mailing list
> [email protected]
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informat=
ics.conductor

_______________________________________________
Bioconductor mailing list
[email protected]
https://stat.ethz.ch/mailman/listinfo/bioconductor
Search the archives: http://news.gmane.org/gmane.science.biology.informatic=
s.conductor