population history
Lohse K, Barton NH, Melika G & Stone GN 2012 A likelihood-based comparison of population histories in a parasitoid guild. Mol Ecol 21:4605-4617.
- if we want to actually test how concordant spatial histories are between species, we need a statistical, model-based framework
- minimal triplet samples are of course uninformative about the parameters of current populations
- they do contain information about the historical relationships of these populations and are amenable to exact likelihood analysis
- the likelihood of a particular model can be maximised directly from the mutational patterns observed across arbitrary numbers of unlinked loci without loss of information
- even with multiple (10–18) independent loci it is surprisingly difficult to distinguish between simple alternative divergence histories
- this is despite the fact that unlike methods that rely on summaries of the data (summary statistics or gene trees), our likelihood calculation uses all available information
- the historical signal contained in sequence data is inherently limited if histories are young
- despite this, there is no shortage of phylogeographic studies that claim to find signatures of much more complex histories than those we were able to investigate here
- few of these provide statistical tests for the historical scenarios they try to infer
- recent histories are hard (or indeed impossible) to resolve using tens of loci
- when increasing the number of loci by an order of magnitude even recent Pleistocene histories can be inferred
- there is much scope for increasing the realism of model based inference
- analogous expressions for the likelihood of triplet genealogies under more complex models including population size changes, migration and admixture can be derived
- because of the inherent stochasticity of the coalescent, large volumes of data are required to have any hope to distinguish such more realistic models from simpler alternatives in practice
