A new toolbox for hard statistical problems in phylogeny and evolution

Statistical analysis of phylogenetic or evolutionary models – widely used across the life sciences – relies to a large extent on specialized software developed for particular models and inference strategies.

In recent work, we have shown that it is possible to take a completely generic approach to model specification using probabilistic programming, while at the same time supporting automatic generation of efficient inference algorithms.

Specifically, we developed techniques allowing us to generate powerful and flexible sequential Monte Carlo algorithms for arbitrary biological diversification models.

Using this approach, we solved a number of inference problems that had been challenging or impossible to address previously.

In the proposed project, we will expand this work to the full range of phylogenetic models, and a range of new inference methods that have seen little or no use in phylogenetics previously.

The aim is to develop a toolbox allowing computational biologists to easily compose new inference strategies for hard statistical problems in phylogeny and evolution.

In pilot projects with collaborators, we will test the power of our methods in addressing some of the toughest problems in our field.

Specifically. we will be looking at complex diversification models, topology inference, scalable online inference, gene tree-species tree problems, and models that require sampling of character change histories.