antpiron
commented
3 months ago Hi,
I expected to have a diagonal with high log p.val and low everywhere else.
In fact, this is the diagonal...
Methylation sites are probably more than 1 millions and if you do the perfect overlap (same lists for both), the hypergeometric P-value underflows to zero even with the "long double" real number. As a consequence, the P-value is zero for a large patch (red). This is a limitation of the way that numbers are represented in computer.. If you use two distinct list, you most probably won't have this artefact.
Anthony.
Hi,
I wanted to do a control test by choosing the same list for "a" and "b" (they are differentialy methylated probes ranked by fold change (log)). I expected to have a diagonal with high log p.val and low everywhere else. Here's the result.
I am not a specialist of this type of analysis so I don't understand why I have this kind of plot, maybe you have more explications.
Thank you !