DMRcate maxdiff and meandiff calculation

Question

[elenasofia.prietoleon]

Hello, I'm conducting a DMP analysis and comparing the results with a DMR analysis. For the first part, I used limma and for the second part I used DMRcate.

As I understand the first step of DMRcate is actually to calculate the individual t tests using the same procedure than limma, and indeed I get the same number of rejected hypotheses.

I also understand that the column maxdiff refers to the max. (in absolute value) of the CpG coefficients included in the identified region (and the average for meandiff). Nevertheless, when I take the CpGs that form one of the regions and extract the max absolute coefficient the values do not match with the column maxdiff in the DMRcate output, same for the average.

Had anyone else encountered the same problem? is it possible that something else is inside of those columns?

Thanks,

Answer 1

[Tim]

Hi, have you ever found a solution to this? I'm noticing the same. For example, if I look at a boxplot of all meandiff values found by DMRcate, it shows that most of the DMRs are hypomethylated. But when I look at all beta values of the CpGs, it shows the expected bimodal distribution. And if I take a simple DMR with only two CpGs and take the average, the values are really off (e.g. DMRcate meandiff = 0.06 and average beta of the two CpGs 0.21)