AF: Small: Fast and accurate computational tools for large-scale evolutionary inference: a phylogenetic network approach

Michigan State University, East Lansing MI

Investigators

Abstract

A grand challenge in science is reconstructing the "Tree of Life", which is the phylogeny, or evolutionary history, of all species on Earth. The notion of a Tree of Life reflects Darwin's view of evolution as "tree-like", where bifurcating speciation results in an ancestral species giving rise to two genetically isolated descendant species. However, recent studies have challenged this view. "Non-tree-like" evolution due to inter-species gene flow - where DNA is shared between species existing at the same time - has significantly shaped the evolution of a far greater diversity of species than was ever thought possible, including humans and Neandertal, mice, and butterflies. In these cases, the phylogeny cannot be described by a tree, but is instead a more general structure known as a phylogenetic network. Our understanding of evolution and biology is at a crossroads. How frequently is the traditional assumption of tree-like evolution violated in the Tree of Life? What is the evolutionary role of gene flow? Applications include understanding the spread of antibiotic resistance among bacteria, which costs the U.S. over $35 billion and a loss of 23,000 lives annually, and pesticide resistance in weeds, mice, and other pests, which costs the U.S. billions of dollars annually. Phylogenetics, or the discipline which seeks to reconstruct evolutionary histories using biomolecular sequences and other biological data, can shed new light into these questions. Two ingredients are necessary for phylogenetic reconstruction and analysis: suitable biological data for the organisms under study, and computational methods capable of efficiently and accurately analyzing the data. Today, biological data abounds due to recent biotechnological advances, and large-scale datasets are common. However, computational methods have not kept pace. New computational frameworks are needed for fast and accurate phylogenetic network inference and analysis in the era of "big data". To address these challenges, this project will create new computational frameworks for fast and accurate network-based phylogenetic inference using large-scale genomic sequence datasets and evolutionary analysis of continuous biological data. The new methodologies will be validated using a comprehensive performance study. More broadly, this project will enable student training, scientific outreach, open-source software development, and scientific research that may yield new biological and biomedical discoveries. Phylogenies are typically inferred using computational analysis of biomolecular sequence data, and phylogenetic comparative methods are used for evolutionary analysis of continuous biological data (e.g., trait data). Today, "big data" challenges abound due to rapid advances in sequencing and related biotechnologies. In particular, large-scale datasets with hundreds of genomes are now common. The state of the art of phylogenetic inference therefore faces two critical scalability challenges: (1) the number of organisms in a study, and (2) greater evolutionary divergence reflecting the complex interplay of tree-like and non-tree-like evolution. For discrete sequence data, state-of-the-art methods address the second challenge, but are not scalable beyond inputs with a few dozen genomes; for continuous data, scalable approaches are needed to address the second challenge in the context of phylogenetic uncertainty and adaptive evolution. The proposed research creates new computational approaches that address both challenges for discrete sequence data and continuous data. The first objective is to create a novel computational framework for scalable phylogenetic network inference using large-scale genomic sequence data. The framework makes use of the multi-species network coalescent model to account for genetic drift, incomplete lineage sorting, and gene flow as well as traditional substitution-based models of sequence evolution. The framework builds on the PI's work on large-scale phylogenetic tree inference by adapting divide-and-conquer algorithms to the more general case of networks, resulting in accurate and efficient inference. The second objective is to develop novel stochastic models and methods for analyzing continuous character evolution on phylogenetic networks. The new models will generalize widely-used non-neutral models of continuous character evolution that assume tree-like evolution, and will be used to create new methods for phylogenetic inference using heterogeneous large-scale inputs. The third objective is to validate the new computational methodologies using new empirical and synthetic benchmarks. The empirical benchmarks include mouse, plant, and fungal datasets that have been produced through ongoing collaborations.