Project description
A probabilistic approach to the statistical analysis of phylogenetics
Statistical analysis of phylogenetic models is currently one of the most active research areas in computational biology, with wide applications in the theory of evolution, epidemiology and forensics. Existing computational approaches to inference based on Markov Chain Monte Carlo (MCMC) methods are not so efficient as Sequential Monte Carlo (SMC) inference algorithms. The goal of the EU-funded PhyPPL project is to apply probabilistic programming to automatically generate SMC inference models for those phylogenetic problems difficult to solve with MCMC methods. Specifically, the project will design statistical inference algorithms for complex diversification models with variable tree topology and a trait-dependent branching process. To demonstrate the potential of the new algorithms, the project will use them to trace the impact of the Andean orogeny on neotropical biodiversity.
Objective
Statistical analysis of phylogenetic models is one of the most active areas of research in computational biology today with wide applications in the Theory of Evolution, epidemiology, forensics, etc. Current phylogenetic software packages limit the user to the set of phylogenetic models and inference strategies that have been pre-programmed in the tool. Inference under certain important phylogenetic models is very difficult with the Markov chain Monte-Carlo strategy implemented in current packages for phylogenetic analysis. The new paradigm of probabilistic programming, coming from computational statistics and theoretical computer science, solves the model expression problem and enables the user to implement novel inference methods. We utilize probabilistic programming to automatically generate Sequential Monte Carlo (SMC) inference machinery for MCMC-hard problems in phylogentics. SMC algorithms may be more efficient, provide unbiased solutions, and provide likelihoods estimates for model comparison.
The goal of the proposed research is to carry out some of the first applications of probabilistic programming to real-world problems of empirical interest in evolutionary biology. The objectives are (1) to design and implement statistical inference machinery for complex diversification models with variable tree topology and a trait-dependent branching process under probabilistic programming, (2) to do a pilot study on the applicability of this inference machinery by studying the effect of the orogeny of the Andes on Neotropical biodiversity, and (3) contribute to the design and implementation of a novel probabilistic programming language for phylogenetics, TreePPL, by utilizing the insights gained from (1) and (2).
We also propose dissemination and communication measures that target scientists and the general public throughout Europe and in particular new and aspiring EU member states.
Fields of science
Keywords
Programme(s)
Funding Scheme
MSCA-IF - Marie Skłodowska-Curie Individual Fellowships (IF)Coordinator
SE 114 18 Stockholm
Sweden