Unify logic for dealing with ploidy

[shz9]

The codebase currently has 2 strategies for dealing with ploidy in Pedigrees:

[1] Specify ploidy at the stage of path-sampling, which involves:
[1.1] Assigning number of specified "ploids" to each individual.
[1.2] Sampling a path through those ploids.
[1.3] Adding a flag to the traversal object so we know its ploidy.

Advantages:

• In principle, it can accommodate any ploidy level (1, 2, 4, ...).
• It has data structures (e.g. Haplotype) to assign values and attributes to individual ploids (e.g. IBD segment).
• Seamlessly integrates with the functionality from Pedigree class.

Disadvantages:

• More messy to deal with in the Aligner object. We will need to check that tree sequence to pedigree node assignment doesn't happen more than ploidy level.

/// Alternative to this formulation ///

As an alternative, we can add an attribute to the Pedigree class self.diploid_graph, which we can initialize in path sampling. The only difficulty with this alternative is how to access the Genealogical helper methods, such as .predecessors(), .successors() and .iter_nodes().

---

[2] Create a DiploidGraph class that inherits from Pedigree
[2.1] Takes an initialized Pedigree and converts it into a DiploidGraph.

Advantages:

• Simpler and more elegant way to carry out the alignment for diploids.

Disadvantages:

• Accommodates only haploids/diploids.
• Because it inherits from Pedigree class, some of the methods may break for the diploid or not make sense in this context.

/// Alternative to this formulation ///

Do not inherit from Pedigree. Just set the Pedigree to be an attribute of the DiploidGraph object. This would mean duplicating the .sample_path() method, but it should be fine.

Any thoughts?

[shz9]

For the alternative formulation of the 2nd solution, maybe we can make DiploidGraph inherit from Genealogical instead of Pedigree? Not sure if it would break anything there.

[ivan-krukov]

I don't think that inheriting from Geneological should break anything. At least it shouldn't, in principle

[shz9]

OK. I updated the interfaces for handling haplotype structures.

I re-named the DiploidGraph class to HaplotypeGraph and implemented interfaces for generating graphs for:

• Autosomal segments.
• X Chromosome segments.
• Y Chromosome segments.
• Mitochondrial segments.

I also moved path sampling to be entirely within the confines of the HaplotypeGraph. The new sample_path method provides an option called haploid_probands, which gives you the option to sample 1 haplotype per proband and then sampling the paths from those. As we discussed, this is a more biologically meaningful way of representing the "haploid" graph.

Note that now Aligner objects take a HaplotypeGraph and a Traversal object as arguments, instead of Pedigree and Traversal as it was implemented previously.

Sorry these changes are a bit too much and may have broken other scripts/test that you've written. But I think this formulation is a lot clearer and will hopefully be useful in the tasks ahead.

As always, examples are in the notebook.