Hi everyone, 

I have this paired experiment: control (16) and treated (17) groups, each was submitted to SD and HF diet. 

1) I want to know the effect of SD in control samples?

count matrix contains control samples only and  design = ~ subject.n + Diet where subject.n is the samples origin (patient id:1,1, 2,2..) Shall I add any interaction term to answer this question?

2) I want to know the effect of SD in treated samples? 

count matrix contains treated samples only and design similar to the previous. similarly should I include any interaction term?

3) I want to know the DEG between the treated and the control samples? 

should i remove the unpaired treated sample? what shall be the design in this case? 

design~ condition + diet + subject.n but what should be the interaction terms? 

Thank you for your help.

I would use the same design for (1) and (2). I suppose that the subjects are nested within control or treated groups? If so, you would use ~condition + condition:subject.n + condition:diet to obtain diet effects for each condition group. These will be coefficients you will see in resultsNames(dds) and you can pull them out using results(dds, name="..."). With the design where you control for subject, you cannot compare directly across control and treatment, because you have controlled for subject which is nested within. If you wanted to get a sense for control vs treatment for a given diet, you could make a subset of the dataset with just that diet and run with a design of ~condition. That's how I would approach it, but obviously you can't control for subject for that comparison.