Variable recombination maps e.g. per population

[hyanwong]

At the end of the chat with @davidrasm yesterday he brought up an interesting suggestion which was to be able to change the recombination map as well as the demography during a simulation.

It should be possible to hack this is we want to change the recombination map over time by stopping a simulation, changing the map, then restarting from the roots of the previous simulation. But I think it would require msprime changes to allow different populations to have different maps, which was one idea that @davidrasm put forward.

Even more difficult (and maybe logically meaningless for a backwards simulator?), but also nice would be to change the map on a lineage-specific basis. I dan't really know how to do this, or even quite what it means.

Pinging @benjeffery as he was in on the conversation too

[hyanwong]

(by the way, it may be simplest for @davidrasm to do this in SLiM for the time being, although I don't know if SLiM has a "record_full_arg" type option, which he was using)

[grahamgower]

Conceptually, its not hard to imagine having two distinct lineages with two distinct recombination maps, and also permitting those maps to change over time (using the approach you suggest). But what happens when those lineages exchange migrants?

[hyanwong]

If it's in populations, then the migrant takes the map of the population it is in. If it is for a "lineage", then I assume we mean a lineage at a particular point in the genome, or something? I haven't thought is through properly, TBH.

[grahamgower]

Yeah, operationally the simplest thing to do is to apply the recipient lineage's map when/after migration occurs. But does this make sense biologically? And are there (tractable) alternatives that makes more sense biologically? Or perhaps more importantly, is this simplest behaviour useful to users?

[hyanwong]

You'll have to ask @davidrasm but he was talking about cases where the recombination map had some sort of geographical correlation. It might be that was caused by environmental factors (in which case the population approach would be fine) , or it might be due to inherited genomic features, such as Prdm9 changes, in which case it's a lineage-specific thing,. and maybe not really something that you would expect to do in a neutral simulator (I think? - that's what I was alluding to above)

FWIW, his study organisms were plant pathogens (e.g. fungi).

[jeromekelleher]

I think there's an existing issue discussing this, good to link these together if we could.

[hyanwong]

Ah, I didn't realise, sorry. I think this is https://github.com/tskit-dev/msprime/issues/1095. Is it worth closing this issue and moving the conversation there?

[jeromekelleher]

Best close one other the other, I don't mind which

[hyanwong]

Closing this one.