Hi,

I've a multifactor design (5 factors, each having two possible value (0 or 1)). One of the factor specify the type of samples (tumor, control) and the 4 others are informations about mutations in specific genes. In summary :

         normal   geneA   geneB   geneC   geneD
sample1       0       1       0       1       0
sample2       0       0       1       1       1
sample3       1       0       0       1       0

I want to test the effect of these mutations (factor geneA, geneB, geneC and geneD) taking into account the sample type (factor normal indicating if it's a tumor (=0) or not (=1)). Here's my code :

dds <- DESeqDataSetFromMatrix(countData = counts,colData = design,design = ~ geneA+geneB+geneC+geneD+normal)
dds <- dds[ rowSums(counts(dds)) > 1, ]
dds <- DESeq(dds)
res <- results(dds,contrast=c(geneA,1,0))

Do you think that the design ( ~ geneA+geneB+geneC+geneD+normal ) is correct in this case ? or should I add an interaction term ?

thanks

You can't add a meaningful interaction term here. The interaction term in theory would help you test if the mutations have a different effect in tumor vs normal. But you have no mutations in normal (correct?), so it's not possible to make that comparison. The effect of mutations in normal cannot be estimated from lack of data, so the ratio of (the effect in tumor) / (the effect in normal) is undefined as well.

You can use the design you proposed though. Although, I would also consider dropping the normal samples and looking at differences among the tumor samples alone. I'm not sure what they add here.