View Single Post
Old 04-27-2011, 02:18 AM   #5
Simon Anders
Senior Member
Location: Heidelberg, Germany

Join Date: Feb 2010
Posts: 994

First of all: I don't think the multiple testing problem has anything to do with all this. Why would the usual Benjamini-Hochberg procedure not be applicable when you have a time course?

What you are missing is a null hypothesis. Once you have formulated it I can tell you whether DESeq can test it.

As a preliminary though exercise to get there, try to become clear what you mean, e.g., by "genes [that] were differentially expressed within the 4 different timepoints". No gene will keep its expression level perfectly constant over a time course. If a gene changes, this does not at all mean that it is under circadian control. To demonstrate this, you will need a couple of time points, let's say 8 per patient, taken, say, at midnight and a noon, and then you look for genes which are consistently higher (or lower) at daytime than at night. You test against the null hypothesis that the observed variation is only due to noise and not connected to the sleep-wake cycle. (Actually, this is a bad example, as you may have a hard time finding any such genes. After all, our whole metabolism and hence most biochemical processes in the cell are somehow coupled to this rhythm.)

Again: For any kind of statistical test (not just for DESeq), you have to choose a test statistic. In my example it might be "average expression at daytime" and "average expression by night", and the null hypothesis would be that they are the same and their ratio hence deviates from 1 only due to noise.
Simon Anders is offline   Reply With Quote