Recapitating with non-uniform recombination rates about pyslim HOT 6 CLOSED

I think this makes sense @petrelharp, and can be tried without needing to make any changes internally to msprime. I guess msprime could provide a function remap_coordinates(ts, recombination_map) which does what you describe, if it works. You can then recapitate with the output of remap_coordinates (this is better than doing this internally within msprime.simulate I think, because the interface is already complex enough).

Unfortunately I don't have time to work on this at the moment though.

from pyslim.

Sounds good. I don't have time to do this quite at the moment either, but can populate an issue over at msprime with our notes on this process if someone else wants to start on it.

from pyslim.

This problem should go away entirely after msprime 1.0.

from pyslim.

So far, we haven't been able to see how to do this without major replumbing under the msprime hood.

But, as I write this...

The basic problem is that the recombination breakpoints output by SLiM may not match the possible recombination breakpoints that msprime will use. Because of how msprime works under the hood, they have to. But, we could

1. make a maximally-fine-but-still-discrete recombination map in msprime
2. remap the breakpoint positions in the input tree sequence to their nearest smaller position in the msprime map
3. warn/error if this has moved any breakpoints more than some distance or has caused anything else bad to happen
4. proceed with recapitation

Perhaps steps 2 and 3 would be a method that would take a tree sequence and a recombination map. This should work just fine, moving breakpoints by no more than something very small unless there are regions with zero recombination rate. At first I thought it should be in pyslim because you're doing something not-quite-right with a tree sequence that is mostly useful for SLiM->msprime; but actually you'd run into the same problem if you tried to do msprime.simulate(... from_ts=ts) with a ts that was made with a different recombination map.

What do you think, @jeromekelleher? Does this, vaguely, make sense?

from pyslim.

This has been possible since 1223afd (?)

from pyslim.

This has been possible since 1223afd (?)

True - you can pass in a recombination map - the thing quoted above was saying that it's not always possible to have exactly the same recombination map, for reasons related to discretization.

But - not an issue any more. I'll close this and open an issue saying that we should document this or give an example.

from pyslim.