Future work: weighting HMM matching by mutational probability

[hyanwong]

Too complex to implement for the final paper, but it would probably be quite easy to weight the HMM to account for the probability of different sorts of mutation. I would imagine that we would keep the current weighting of e.g. 5 mutations to 1 recombination, but weight the mutations such that some counted more than a unit contribution, and some counted less, with the mean being 1. Then I think we wouldn't have to tweak the cutoffs again.

We could do this iteratively: we could use the find_problematic ARG as a first pass to estimate the probabilities of each type of SNP mutation, then weight the HMM using those probabilities.

I was motivated to think of this because of the large range of probabilities of the different SARS-CoV2 mutation types in https://academic.oup.com/mbe/article/40/4/msad085/7113660:

We see 40x more C->T mutations than e.g. G->C or C->G.

[jeromekelleher]

I think this would have to go into the HMM implementation itself to work properly, but yes, definitely a worthwhile refinement for the future.

[szhan]

Could easily create some tests in https://github.com/astheeggeggs/lshmm before implementing? I think we briefly talked about implementing this extension on and off before.