CpG mutation models

[hyanwong]

Elevated mutation rates at CpG dinucleotides are one of the major contributors to mutation rate variation in mammals. These can't easily be simulated by sim_mutations, but there are probably reasonable approximations if a reference sequence is defined.

@petrelharp says, on slack:

It'd take some thinking to figure out how to do it - since we can't do real context-dependence, the most obvious thing is a dinucleotide model, but then this misses half the CpG pairs. I think a pretty good approximation to real context-dependence could be done by just using the reference sequence when you don't know what the neighboring nucleotide is, but getting that into msprime would be a significant project.

The main time CpG is mentioned on the msprime GitHub repo is in https://github.com/tskit-dev/msprime/issues/972, but only in passing, so I have opened this issue for people who are searching for the term in conjunction with msprime, and who might want to think about an implementation.