Msprime: Update mutation code once mutations have `time`

[petrelharp]

As discussed in #946, this would make adding mutations to different time chunks of a tree sequence really possible, and would be a good idea more generally.

[jeromekelleher]

Needs to get added to tskit first, https://github.com/tskit-dev/tskit/issues/513

[petrelharp]

whoops, sorry, wrong repository. =) But this will remind us to update the mutation code.

[jeromekelleher]

Waiting on #1114

[jeromekelleher]

We now have a time column, anyone want to fill it in? Should be a straightforward update.

[mufernando]

I'll give it a go.

[brianzhang01]

Hello -- chiming in with a particular question, and I hope this is an appropriate thread for it. I haven't followed all the discussions super closely but just caught up now.

I've been playing with the mutation time feature, which is awesome, and right now I've taken a tree sequence, sampled a lot of mutations, and plotted the distribution over times. Since msprime gave me unknown times, I used the following code:

temp_ts = msprime.mutate(simulation, rate=args.mu*10, random_seed=seed)
temp_tables = temp_ts.dump_tables()
temp_tables.compute_mutation_times()
temp_ts = temp_tables.tree_sequence()
mutation_times = np.array([m.time for m in temp_ts.mutations()])
histogram = np.histogram(mutation_times, interval_bins, density=False)[0]

However, I was surprised that compute_mutation_times() currently gives the midpoint time from the node below and node above, rather than a uniform sampling over the height of the branch. Am I correct in guessing that uniform sampling, because it involves randomness, is instead a feature for msprime? This is really exciting, thanks @mufernando and everyone else for working on it.

[benjeffery]

@brianzhang01 thanks for reminding me of this. We've discussed this issue in the original mutation time work: https://github.com/tskit-dev/tskit/issues/513#issuecomment-639539212 The end result was to add as a future feature sampling uniformly, and getting round the need for random-ness by passing in a random array. However it seems there are other uses for randomness (e.g. https://github.com/tskit-dev/tskit/pull/815#issuecomment-682989666) so I'm starting to think we bite the bullet and solve the MIT-compatible source of randomness problem.

Long story short - this is planned and I should have made an issue! (thought I had but can't find it!)

[benjeffery]

Issue now at https://github.com/tskit-dev/tskit/issues/849

[brianzhang01]

Thanks @benjeffery for catching me up! So it seems like on the tskit side, you'll be working on the pseudorandom uniform sampling in compute_mutation_times(), and this issue will also add pseudorandom uniform sampling on the msprime side? (simulate / mutate / sim_mutations) For me, I would be happy if just one of these were in (I'll roll my own in the meantime using the Python API). I'll also comment on the tskit issue you linked.

[jeromekelleher]

Closing this as times supported since #1179.