We can look at the reconstructed ARG and tell
1) how many back-mutations (or recurrent mutations) we get
2) how many identical recombination events we have (at the same place, with the same two ancestors on either side).
if we have large numbers of either, it implies that we have done badly at reconstruction. We could also use this measure to work out what the best mutation:recombination parameter to set for a given case - it is the one which minimises these two measures.
If our reconstructed ARG comes from simulated data, we can also calculate how many real events like (1) and (2) we had in the simulation, which gives us a handle on the expected number. I suspect (2) should basically be 0, whereas (1) will be seen a handful of times.
Copied from original issue: mcveanlab/treeseq-inference#40
From @hyanwong on May 17, 2017 10:2
We can look at the reconstructed ARG and tell 1) how many back-mutations (or recurrent mutations) we get 2) how many identical recombination events we have (at the same place, with the same two ancestors on either side).
if we have large numbers of either, it implies that we have done badly at reconstruction. We could also use this measure to work out what the best mutation:recombination parameter to set for a given case - it is the one which minimises these two measures.
If our reconstructed ARG comes from simulated data, we can also calculate how many real events like (1) and (2) we had in the simulation, which gives us a handle on the expected number. I suspect (2) should basically be 0, whereas (1) will be seen a handful of times.
Copied from original issue: mcveanlab/treeseq-inference#40