Improved forecasts and visualization of seasonal influenza evolution

Abstract

The rapid evolution of seasonal influenza requires the development of new vaccines every one to two years. This evolution occurs through a process of antigenic drift where amino acid mutations in the hemagglutinin surface protein allow currently circulating viruses to evade adaptive immunity against previous viruses. Vaccine composition decisions are guided by predictions made from serological assays of antigenic drift and sequence-based forecasting models. These predictions do not account for functional effects of mutations measured by deep mutational scanning experiments or attempt to integrate fitness effects measured by experimental and sequence data. In this dissertation, I attempted to understand whether experimental measurements of antigenic drift and functional constraint could be used to improve forecasts of seasonal influenza evolution. I found that most estimates of seasonal influenza fitness could not robustly forecast future populations. Models that integrated serological measurements of antigenic drift with sequence-based estimates of functional constraint provided the most robust forecasts. I concluded that successful seasonal influenza predictions depend on the choice of prediction targets and fitness metrics.