How DESeq2 formula correction works?

Question

[Rafael Soler]

Hello,

I have some doubts after reading the DESeq2 vignette and different threads on this topic. If I have this data:

If my objective is to compare the Status of the samples, in principle the formula would be ~Status. However, there are big differences between the Zone, as we can see in this PCA:

Would we have to make the formula like ~Zone+Status in order to control the effect of the Zone over Status? Or is it better to do it only with ~Status? If a new variable such as Sex is added and we want to control its effect, we would have to reconvert the formula to ~Zone+Sex+Status? Where can I find an explanation of what this "control" is based on?

Thank you

[brunobsouzaa]

I would rely on how much the Zones are affecting your Status. Did you try doing Status ~ Zones and checking the results? If you see that your Zones are really influencing your Status, then do ~ Status + Zones. Same thing for a new variable Sex as you mentioned.

[Rafael Soler]

Thanks. In the DESeq2 vignnete we can see that design = ~ batch + condition. Shouldn't it be in this order? design = ~ Zones + Status

Best

Answer 1

[swbarnes2]

Use zone + status.

[Rafael Soler]

Why its better to use this formula rather than only ~Status? If a new variable such as Sex, we would have to reconvert the formula to ~Zone+Sex+Status?

Thanks for the reply, although I would appreciate a bit of explanation.

Best

[swbarnes2]

You have variability within each group caused by zone. If you omit zone, the software just thinks there is a lot of variability. If you include zone, then it understands that there is an underlying reason for some of the variability and it can be included in the mathematical model, so you get better results.

And yes, you can just pile sex, or batch in the design like that.

[Rafael Soler]

Thanks, now I understand it better.

[swbarnes2]

Since Zone is such a huge contributor to variance in your data, I'd run it with each zone alone.