search

hyanwong / molecular-sample-dating

Methods for dating ancient samples using tree sequence methods
0 stars 0 forks source link

Evaluate the accuracy of inferring tree sequences with ancients #2

Open awohns opened 1 year ago

awohns commented 1 year ago

@awohns has tried a new approach to inferring tree sequences with ancients, where all samples (modern and ancients) are simply added to a SampleData file and the tree seq is inferred. This appears to work better than the iterative approach.

Two ways to evaluate:

  1. Toplogical inference accuracy (KC distance, RF etc.)
  2. Date estimation: does the tree sequence allow for better dating than the iterative approach?
hyanwong commented 1 year ago

The pipeline for dating in one step including historical samples is at https://github.com/hyanwong/molecular-sample-dating/blob/main/notebooks/one_step_ancient_pipeline.ipynb

hyanwong commented 1 year ago

One issue with this pipeline is that the conditional coalescent prior is inaccurate in this case, as it will be assuming that all the samples are at time 0. This shouldn't be an issue if most of the samples are recent, but if most of the samples are ancient, the prior will be wrong, I think.

hyanwong commented 1 year ago

We should also check if we do better with a lognormal or a gamma prior in this new pipeline.