Good morning.

I have something that I am confused about when trying to use DESeq2 for my analysis. I read the tutorial from the developer of DESeq2, but I just wanted to confirm if I am understanding how LRT works. In this scenario, I have a dataset with conditions (control, cases). I want to test the differential gene expression while controlling for age and gender. This is the code that I have:

dds <- DESeqDataSetFromMatrix(countData = round(df),
                                      colData = meta,
                                      design= ~ condition + gender + age)

dds <- DESeq(dds, test = "LRT", reduced = ~ gender + age)

Would it be correct in saying that when I leave out "condition", I am testing to see if there is a significant impact from "condition" while controlling for age and gender? In other words, I am a bit confused on what should or should not be in the reduced formula.

That's how you should do it.

The log likelihood is a measure of how well the model fits the data, and the likelihood ratio test compares the log likelihoods for the full and reduced model to say if the additional covariate(s) improves the model fit significantly. In this case you are asking if including the case/control status in your model improves the fit over the model with just gender and age.

You always include any nuisance variables (like age and gender) in the reduced model, because you are assuming that you need them for at least some of the genes.