Distance Aware Algorithms For Scalable Evolutionary And Ecological Analyses

Thanks to the advances in sequencing technologies in the last two decades, the set of available whole-genome sequences has been expanding rapidly. One of the challenges in phylogenetics is accurate large-scale phylogenetic inference based on whole-genome sequences. A related challenge is using incomplete genome-wide data in an assembly-free manner for accurate sample identification with reference to phylogeny. This dissertation proposes new scalable and accurate algorithms to address these two challenges. First, I present a family of scalable methods called TreeCluster for breaking a large set of sequences into evolutionary homogeneous clusters. Second, I present two algorithms for accurate phylogenetic placement of genomic sequences on ultra-large single-gene and whole-genome based trees. The first version, APPLES, scales linearly with the reference size while APPLES-2 scales sub-linearly thanks to a divide-and-conquer strategy based on the TreeCluster method. Third, I develop a solution for assembly-free sample phylogenetic placement for a particularly challenging case when the specimen is a mixture of two cohabiting species or a hybrid of two species. Fourth, I address one limitation of assembly-free methods--their reliance on simple models of sequence evolution--by developing a technique to compute evolutionary distances under a complex 4-parameter model called TK4. Finally, I introduce a divide-and-conquer workflow for incrementally growing and updating ultra-large phylogenies using many of the ingredients developed in other chapters. This workflow (uDance) is accurate in simulations and can build a 200,000-genome microbial tree-of-life based on 388 marker genes.