Threshold for Differentially Methylated Regions

Question

[niutster]

Hi,

Which threshold is appropriate for determining meaningful differentially methylated regions in biology? I mean, which degree of methylation difference between two tissue is meaningful ?

[ivivek_ngs]

I would not generalize it. Ideally, thresholds depend on selection criteria one would like to use a cut-off. You can allways plot the distribution of the delta values and view how it is. Then select a criterion based on that , stating something which is hyper or hypomethylated. Based on the distribution it is ideally suited to chose the values be it, 0.2, 0.3 or 0.5. However the first selection should be based on your FDR or p.value and log2FoldChange. I would expect you are performing differential methlyation with MACS2 or some other standard tools which usually takes the edgeR mode and so you should have the log2FC and FDR/p.value criterions with you. You can make the first selection based on them and then put the additional threshold of delta values based on its distriubtion to make it more stringent of relaxed.

Answer 1

[mac]

Delta beta values greater or equal to 0.2 are generally considered to signify differential methylation but as with most cut-offs it is somehow arbitrary.

Answer 2

[Aerval]

You cannot set a definitive threshold when change in methylation is meaningful. Sure, a single cell is always 1, 0.5 or 0 methylated at one site, but you have bulk data from probably thousands of cells. Statistical tests are for sure the way to go but differ (depending on your type of data and some underlying assumptions) between DMR callers (which also leads to different numbers of DMRs being called for each of them). So you can obviously not set a general threshold, a long DMR of 30 CpG (or much more) might have a much lower difference than a short one but might nevertheless be more interesting. Also, your input material might not be homogenous and contain 10% of a differentially methylated population, so 0.1 (or much smaller) might be interesting here or there. Its true though that we usually use the most differentially methylated regions by difference.