Calculating inbreeding coefficients from pedigrees

[JimmyPeniston]

Hi all,

I have tree sequence data from a pretty simple non-WF SLIM simulation and I'm wondering if anyone has advice for a straightforward way to calculate pedigree-based inbreeding coefficients. The best method I can think of is to try building some sort of recursive algorithm that loops through individuals.parents, but I'm not sure if there is an easier/more efficient way.

For context, I am currently only interested in pedigree-based metrics. So, my chromosome length is set to 1 and a mutation rate of 0.

Thank you for your help. Sorry if this is a silly question or the wrong place to post this. I'm very new to using tskit.

[jeromekelleher]

Hi @JimmyPeniston 👋

We don't have any code for directly computing pedigree statistics in tskit, but there are some very powerful methods in sgkit like pedigree_inbreeding which will hopefully do what you need.

The sgkit code has just been released and I don't think anyone has linked up simulations with it yet. I guess the main thing is whether SLiM saves the pedigree information in the individual table (which I'm not sure about).

Can you run a very small example, and print out the contents of the individual and node tables here please (e.g. print(ts.tables.nodes))? Hopefully hooking the tskit data to sgkit will then be fairly straightforward.

This is probably quite a useful thing in general, so we should document how to do it somewhere, I guess (assuming it works).

[jeromekelleher]

Thanks @timothymillar, that's exactly what I was hoping to do ❤️

[JimmyPeniston]

This is great! Thanks @timothymillar @jeromekelleher @petrelharp!!!! You have all been incredibly helpful.

[jeromekelleher]

Would you mind reporting back on how you get on using the code @JimmyPeniston? It would be useful to know how well it scales for these types of sims.

[JimmyPeniston]

Certainly! I'll report back here with more updates once I've expanded the code for what I need.
As of now, I just tried running the above code for a simulation with 10,000 stored individuals and it took less than 5 seconds.

[timothymillar]

Just a note on the pedigree_inbreeding implementation; It uses a (unrolled) recursive algorithm to calculate a minimal set of kinship coefficients required to get the inbreeding coefficients. The performance depends upon the complexity of the pedigree rather than just the number of samples. I haven't gone through the exact complexity notation yet but a shallower pedigree, with more parent reuse, will be much faster than a deep pedigree with random parents.