Differentially expressed genes in response to a measured clinical parameter

[sindrle]

Hi!
I don't know if I got this right, but is there a way, using DESeq/edgeR, getting diff.exp. genes related to a measured parameter, lets say glucose oxidation change.

If you have two time points, and glucose oxidation changes significantly. You measure this as a continuous variable. Can you somehow add this information and thus get diff.exp. genes related to glucose oxidation?

Thanks!

[dpryan]

As long as the change isn't always partitioned (i.e., divided completely) according to the timepoint then you can probably do that. You can use continuous variables instead of factors in both DESeq(2) and edgeR. There might be an example of that in the edgeR vignette, I don't exactly remember. Either way, keep in mind that you may want to log or log10 transform your continuous covariate, depending on its range.

[sindrle]

As long as the change isn't always partitioned (i.e., divided completely) according to the timepoint

Glucose uptake will gradually increase from Time 1 - Time 2, due to the intervention. So I guess thats fine?

keep in mind that you may want to log or log10 transform your continuous covariate, depending on its range.

Thanks, but why would I want to do that?

[dpryan]

Glucose uptake will gradually increase from Time 1 - Time 2, due to the intervention. So I guess thats fine?

I guess it'll depend on how heavily glucose uptake correlates. The higher the correlation the less the model will fit things, since it'll become unclear whether a change is due time or glucose uptake. In that situation it's best to just ask what's more important to the questions at hand (I'd guess glucose uptake).

Thanks, but why would I want to do that?

It's mostly a question of units. Remember that fitting data with a model is asking it to find a coefficient describing a change for a 1 unit increase in whatever you're measuring. So, if your glucose changes are large on some scale, then you can run into numerical stability issues.

[sindrle]

Oh, I see. Thanks for pointing that out!