A geometric law for DNA methylation across 82 human cell types

Jan Fabio Nickels, Postdoctoral Research, Center for Quantitative Analysis of Molecular and Cellular Biosystems (BioQuant), Heidelberg University, Heidelberg, Germany

DNA carries a chemical mark, a methyl group, at one particular two-letter sequence: CpG. Which CpGs are marked is much of what distinguishes one human cell type from another, and the pattern is copied every time a cell divides.
Along a chromosome the marking is all-or-none: a stretch of DNA is either almost entirely marked or almost entirely bare, and stretches in between are rare. That CpG-dense regions tend to be the bare ones has been known since the 1980s. Yet how the two ingredients trade off has not been — how the chance of staying bare depends on how many CpGs a stretch holds, and on how far apart they lie.
Measuring this across 82 sorted human cell types, I find that the two combine into a single quantity, n/L^q. Density would mean q = 1. The measured exponent is 0.567, and it leaves 6.5 times less variance unexplained than density does. Since q is below one, spreading CpGs apart costs less than density says it should: they are not acting one at a time, but together across the whole stretch.
I will show how the exponent was measured, what I did to try to break it, and where the differences between cell types sit in this picture.