Hi List, I am having a bit trouble pinning down the interaction terms in DESEQ2. I have followed the user manual, read previous q&a's, and thought I understood. But upon actual application, I realized that I am still confused. For simplicity reasons, I will just ask my questions based off the example in user manual 1.5, under section 3.3. >The main effect of OHT treatment over control for all patients can be extracted with the following: >resOHT <- results(ddsX, contrast = c("treatment","OHT","Control")) check. >In order to test if the effect for OHT treatment over control is different for patient 4, we test the interaction term with the folloing: resX <- results(ddsX, contrast = list("patient4.treatmentOHT", "patient4.treatmentControl")) 1) Point of clarification, effect for OHT treatment over control is different for patient 4 (over all other patients?) As in an ANOVA comparison? These genes are only differentially expressed b/w OHT and control in patient 4? 2) This is where I am most confused. If there are only 2 levels in the patient variable, patient 4 and patient 5, I'd thought that this would be equivalent to: contrast=list("patient4.treatmentOHT","patient5.treatmentOHT") But it's not. So what does the above contrast mean? 3) if we want to do this example with LRT, do we need to fit with a different reduced model depending on what contrast we want to pull out? Thanks in advance. -Rob [[alternative HTML version deleted]]

Hi Rob On Jul 30, 2014 4:54 PM, "rob yang" <nextgame@hotmail.com> wrote: > > Hi List, > I am having a bit trouble pinning down the interaction terms in DESEQ2. I have followed the user manual, read previous q&a's, and thought I understood. But upon actual application, I realized that I am still confused. > > For simplicity reasons, I will just ask my questions based off the example in user manual 1.5, under section 3.3. > > > > > > > > > > > > > > > >The main effect of OHT treatment over control for all patients can be extracted with the following: > > > > > > > > >resOHT <- results(ddsX, contrast = c("treatment","OHT","Control")) > > > > > > > check. > > > > > > > > > > > > > >In order to test if the effect for OHT treatment over control is different for patient 4, we test the > interaction term with the folloing: > > > > > > > > > > > > > > > > > > > resX <- results(ddsX, contrast = > list("patient4.treatmentOHT", > > "patient4.treatmentControl")) > > > > > > > 1) Point of clarification, effect for OHT treatment over control is different for patient 4 (over all other patients?) As in an ANOVA comparison? These genes are only differentially expressed b/w OHT and control in patient 4? > > The interaction term means you test for the hypothesis that each patient has its own effect. These genes are DE b/w OHT and control different in patient 4 beyond the general effect. > > > > > > > 2) This is where I am most confused. If there are only 2 levels in the patient variable, patient 4 and patient 5, I'd thought that this would be equivalent to: > contrast=list("patient4.treatmentOHT","patient5.treatmentOHT") > But it's not. So what does the above contrast mean? > This contrast doesn't have control in the denominator. I don't see why it should be equivalent. > 3) if we want to do this example with LRT, do we need to fit with a different reduced model depending on what contrast we want to pull out? > In DESeq2, the LRT only tests all interaction terms. Best, Mike > Thanks in advance. > > -Rob > > > > [[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor@r-project.org > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor [[alternative HTML version deleted]]