Hi Everyone!

I am using dmpFinder to find dmp on a methylation experiment.

As the authors said, dmpFinder is a wrapper around limma, so I tried to emmulate the steps (that I thought) were followed by the dmpFinder function... After examining the function @HansenLab's Github I found a series of steps that I don't fully understand...

Is anyone able to explain the statistics behind the dmpFinder function in minfi package? Why does dmpFinder applies bayes in some cases, or else, uses Sigma?? (Please, follow the link to Hansen's Lab to see the code I am talking about...) Could someone, please, guide me on why these steps are followed? or what do they mean?

I am sorry if the question is too basic, maybe it is a matter of the limma package, but I can't seem to find the answer.

Many thanks in advance.

https://github.com/kasperdanielhansen/minfi/blob/master/R/dmr.R

Hi James.

Many thanks for you reply.

Maybe I should give you more background.

I am trying to select an N amount of probes that are differentially methylated based on a linear model (for further analysis that talking about it doesn't fit in this question)

Since I have a constraint to choose N amount and not just all the data coming from the 450k experiments, my first shot was to preselect them based on the dmpFinder outcomes. But later, I learned that the function is just a wrapper around limma. So I tried with limma to see if got the same dmp.

With “minfi” package, to get those dmp, I do:

as you can see I don’t use ('shrinkVar=TRUE'), because I know my sample size is not too small…

However, I don’t get the same result if I apply the following with limma:

I can see I get a different result because, since I didn’t choose to ('shrinkVar=TRUE'), then the code that is running is the following:

So, I’d like to know what lies beneath (statistically speaking) this piece of code. Because my understanding is that limma has a straight forward way of being used. Unless, it has to be used differently depending on …?(?)based on your answer, depending on the number of samples, I suspect???

If so, then I’d like to know the statistics that drive those steps, so that I can apply the same logical steps without just copy-pasting what has been done for the dmpFinder.

Many thanks,