Forecasting H1N1 Influenza Pandemic and Seasonal Evolution

The rapid accumulation of mutations in influenza strains challenges vaccine efficacy by enabling immune escape. Current influenza vaccine strategies rely on on annual updates that incorporate recently circulating strains, with strain selection guided by experimental neutralization assays and global surveillance data. However, these methods are resource-intensive and have limited predictive power, often resulting in mismatches between the strains included in the vaccine and the strains dominant at the time of vaccine rollout --- ultimately reducing vaccine effectiveness. Understanding how specific mutations alter viral fitness and antigenicity could improve our ability to forecast emerging escape variants and design more effective vaccines. To address this, we adapt EVE, a deep generative model, and retrain it on influenza hemagglutinin (HA) sequences to model the evolution of H1N1. Remarkably, by training on HA sequences sampled prior to the emergence of a given clade, our model consistently predicts the most frequent future mutations across all seven major H1N1 clades since the 2009 pandemic—without requiring prior knowledge of clade identity. This work highlights the potential of deep learning models to forecast influenza evolution and support proactive vaccine design.