Statistical Innovation to Integrate Sequences and Phenotypes for Scalable Phylodynamic Inference

PROJECT SUMMARY/ABSTRACT This proposal targets the design, development and distribution of Bayesian statistical methods and software to study the historical and real-time emergence of rapidly evolving pathogens, such as Dengue, Ebola, hepatitis B, HIV, influenza, mpox, SARS-CoV-2, and yellow fever viruses. The proposal exploits novel versatile and scalable modeling and inference techniques to equip us for large-scale epidemics and pandemics and address some of the pressing and long-standing questions in viral epidemiology. Our multidisciplinary team carries expertise across statistical thinking, data science, evolutionary biology and infectious diseases to leverage advancing se- quencing technology and high-throughput biological experimentation that can characterize 10,000s of pathogen genomes, phenotype measurements or covariates, including sampled geographic, epidemiologic, ecologic and clinical information, from a single outbreak, to help inform actionable public health policy. Our chief innovations are three-fold. First, we will invent versatile phylogenetic and phylodynamic models to better understand com- plex biological heterogeneity in the emergence and spread of rapidly evolving pathogens within and across host reservoirs. Second, we will foster scalable phylodynamic techniques to more accurately learn the dynamics of pathogen transmission between infected individuals and the effects of population structure. Third, we will de- sign markedly more time- and energy-efficient Bayesian phylogenetic software that exploits recent advances in artificial intelligence (AI) techniques and massively parallel computing. Although no alternative implementations exist for the phylogenetic, phylogeographic and phylodynamic models we are developing at this scale, we will compare restricted cases of our models with reduced datasets to current state-of-the-art approaches to evaluate computational performance improvement and statistical bias that these limitations inject using real-world exam- ples. We will demonstrate the value of our developments through high-impact applications across a network of on-going collaborations. These include testing for seasonal persistence of yellow fever virus and improving deep history reconstruction of hepatitis B through novel time-inhomogeneous evolutionary models, and measur- ing the impact of control strategies on Dengue virus and identifying the ecological drivers of highly-pathogenic avian influenza spread while mitigating persistent sampling bias. This proposal will deliver user-friendly software through a leap-forward version of the popular BEAST platform for deployment across a rapidly expanding range of large-scale problems in the molecular epidemiology of infectious disease.