add dog demographic model

[agladstein]

I am making this a draft because I have a few unresolved issues.

• I cannot find the Boxer Ne
• When I try to build the docs locally, I do not see the demographic model update to the catalog
• Also, I think the ancestral populations probably shouldn't be "sampled". I don't know how to do that.

This PR will close #830

[codecov]

Codecov Report

Merging #905 (3a1d76f) into main (541d561) will decrease coverage by 2.57%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##             main     #905      +/-   ##
==========================================
- Coverage   99.40%   96.83%   -2.58%     
==========================================
  Files          80       81       +1     
  Lines        2521     2588      +67     
  Branches      285      285              
==========================================
  Hits         2506     2506              
- Misses          7       74      +67     
  Partials        8        8              

Impacted Files	Coverage Δ
stdpopsim/catalog/CanFam/demographic_models.py	0.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 541d561...3a1d76f. Read the comment docs.

[agladstein]

The Boxer Ne is not inferred, so I'm not going to include the Boxer in the model.

[apragsdale]

Hi @agladstein - looks like it's very close. Just commenting that the mutation rate can now be added to the DemographicModel as that PR was merged, and a question about the mutation rates.

[apragsdale]

The mutation rate can now be added to the DemographicModel.

[apragsdale]

Some of these seem like very high mutation rates.. are they unscaled?

[petrelharp]

If you can finish this up we could get it QC'ed, probably, @agladstein!

[agladstein]

Ok, I'll try to do it this weekend.

[agladstein]

I started to work on this, but I'm so out of it (computer isn't setup anymore, weird merge conflicts...), I won't be able to finish it this weekend.
Can anyone else pick this up to get it done?

[petrelharp]

I'll ask on slack!

[rwaples]

I can give this a try.
At a first glance it looks almost done - I'll see if can tackle the issues @agladstein mentioned at the top.
My initial plan would be paste the changes over to a new fork. I think that will be easier for me rather than trying to pick up this merge. Does that seem like a bad idea?

[agladstein]

That seems reasonable to me @rwaples. I got blocked over the weekend because of merge conflicts, so that sounds like a reasonable way around that problem.
I think it is almost done! I just don't have the bandwidth right now.

[rwaples]

Sounds good - you did almost all the hard work.

I'd be curious what your thoughts on including the Boxer population.
Based on my reading of the paper, they don't seem to have any samples from the boxer, just the haploid dog genome, which is from a Boxer.
As you noted they don't seem to report any population sizes for the Boxer, but they do give a split time between Basenji and Boxer, as well as an ancestral (Boxer, Basenji) population size.

I see a few options:

1. Give Boxer an arbitrary size (e.g. 1000), as you have in your draft PR
2. Give Boxer a size of 1, in effect signaling we don't know
3. Give Boxer a population size of 0 - either explicitly or implicitly removing it from the model, so that it cannot be sampled from.
4. Ask the paper authors if they can help
   Any thoughts?

[petrelharp]

Here's some options I can think of:

1. remove Boxer from the model
2. set the population size equal to the parent population size (Boxer-Basenji?)
3. pull an effective population size from some place else (maybe here?)

IMO, any of these would seem reasonable and so would be fine, as long as it's documented what's happened.

[agladstein]

I have edits locally that deal with boxer, so apparently me a year ago had decided on something (but merge conflicts making it difficult). I just removed Boxer from the model.

[rwaples]

Thanks for the input.
Removing the Boxer makes sense to me, as it doesn't seem very appropriate to use this model to simulate Boxer.