TreePPL: A Universal Probabilistic Programming Language for Phylogenetics

Abstract

ABSTRACTWe present TreePPL, a language for probabilistic modeling and inference in statistical phylogenetics. Specifically, TreePPL is a domain-specific universal probabilistic programming language (PPL), particularly designed for describing phylogenetic models. The core idea is to express the model as a computer program, which estimates the posterior probability distribution of interest when executed sufficiently many times. The program uses two special probabilistic constructs:assumestatements, which describe latent random variables in the model, andobservestatements, which condition random variables in the model on observed data. Theassumeandobservestatements make it possible for generic inference algorithms, such as sequential Monte Carlo and Markov chain Monte Carlo algorithms, to identify checkpoints that enable them to generate and manipulate simulations from the posterior probability distribution. This means that a user can focus on describing the model, and leave the estimation of the posterior probability distribution to TreePPL’s inference machinery. The TreePPL modeling language is inspired by R, Python, and the functional programming language OCaml. The model script can be conveniently run from a Python environment (an R environment is work in progress), which can be used for pre-processing, feeding the model with the observed data, controlling and running the inference, and receiving and post-processing the output data. The inference machinery is generated by a compiler framework developed specifically for supporting domain-specific modeling and inference, the Miking CorePPL framework. It currently supports a range of inference strategies, including several recent innovations that are important for efficient inference on phylogenetic models. It also supports the implementation of novel inference strategies for models described using TreePPL or other domain-specific modeling languages. We briefly describe the TreePPL modeling language and the Python environment, and give some examples of modeling and inference with TreePPL. The examples illustrate how TreePPL can be used to address a range of common problem types considered in statistical phylogenetics, from diversification and co-speciation analysis to tree inference. Although much progress has been made in recent years, developing efficient algorithms for automatic PPL-based inference is still a very active field. A few major challenges remain to be addressed before the entire phylogenetic model space is adequately covered by efficient automatic inference techniques, but several of them are being addressed in ongoing work on TreePPL. We end the paper by discussing how probabilistic programming can support the use of machine learning in designing and fine-tuning inference strategies and in extending incomplete model descriptions in phylogenetics.