Methods for quantifying selection and predicting evolutionary dynamics

PROJECT SUMMARY Natural selection is central to many challenges in biology and medicine, from the emergence of drug resistance in pathogens to cancer evolution. Understanding selection can also aid in protein engineering and help identify clinically relevant mutations in human disease genes. Temporal genetic data - sequences and phenotypes sampled over time - can be an especially powerful tool for understanding selection because it allows us to observe evolutionary dynamics directly. But while temporal data from sources like pathogen surveillance, ancient DNA, and experimental evolution have grown tremendously in recent years, statistical analyses of these data remain challenging. My lab will continue to pioneer the development of new computational methods to learn from temporal genetic data, revealing variants and phenotypes under selection and harnessing this information for predictive models of evolution. Over the past five years, we have developed several approaches to quantify selection from temporal data. Thanks to the use of mathematical methods from statistical physics, our methods are fast and accurate despite the inclusion of complex features such as linkage disequilibrium, epistasis, and time-varying selection. We demonstrated the power of these approaches through studies of HIV-1 immune escape and SARS-CoV-2 adaptation during the COVID-19 pandemic, where our analysis identified key mutations affecting viral transmission even before their importance was validated experimentally. Building on this foundation, we will pursue three synergistic research directions: First, we will develop new methods to jointly analyze selection on both individual mutations and phenotypic traits, fusing concepts from population genetics, quantitative genetics, and machine learning. Second, we will apply these methods to study rapid evolution in viral pathogens. Phenotypic models will help us to understand how immune pressure drives antigenic change in respiratory viruses and to compare evolutionary constraints on pathogens across host species. As an ambitious new direction, we will leverage these insights to develop predictive models of pathogen evolution, with influenza as a first target. Our research will systematically identify the features with the greatest power to predict evolution and characterize how and why predictive power may decline over time. Finally, we will extend our approaches to improve the interpretation of high-throughput mutagenesis experiments that measure the effects of thousands of mutations simultaneously. The proposed research will transform our understanding of how selection guides evolution across biological scales, from individual mutations to complex phenotypes, with applications ranging from predicting viral evolution to protein engineering. These advances could ultimately improve our ability to anticipate and control evolutionary processes across a wide range of biological contexts.