Model choice using Approximate Bayesian Computation and Random Forests: analyses based on model grouping to make inferences about the genetic history of Pygmy human populations


  • Arnaud Estoup
  • Louis Raynal
  • Paul Verdu
  • Jean-Michel Marin


In evolutionary biology, simulation-based methods such as Approximate Bayesian Computation (ABC) are well adapted to make statistical inferences about complex models of natural population histories. Pudlo et al. (2016) recently developed a novel approach based on the Random Forests method (RF): the ABC-RF algorithm. Here we present the results of analyses based on ABC-RF to make inferences about the history of Pygmy human populations from Western Central Africa from a microsatellite genetic dataset. A noticeable novelty of the statistical analyses presented here is the application of ABC-RF methodology to make model choice on predefined groups of models. We formalized eight complex evolutionary scenarios which incorporate (or not) three major events: (i) whether there exists an ancestral common Pygmy population, (ii) the possibility of introgression/migration events between Pygmy and non-Pygmy populations, and (iii) the possibility of a change in size in the past in the non-Pygmy African population. We show that our grouping approach allows disentangling with strong confidence the main evolutionary events characterizing the population history of interest. The selected final scenario corresponds to a common origin of allWestern Central African Pygmy groups, with the ancestral Pygmy population having diverged, with asymmetrical genetic introgression, from a demographically expanding non-Pygmy population.






Special Issue on Models and Inference in Population Genetics