Lin Wang (University of Cambridge)

Speaker: Lin Wang (Pathogen Dynamics Unit, Department of Genetics, University of Cambridge)

Title: Characterising antigenic evolution of influenza A viruses by pathogen genomics and deep learning

Abstract:

Seasonal influenza remains a major public health challenge. The antigenic properties of influenza viruses evolve under immune-driven selection, necessitating frequent vaccine updates. Currently, characterising antigenic drift relies on labor-intensive laboratory work, including ferret inoculations and hemagglutination inhibition (HAI) experiments, which typically provide a six-month lead time for predicting the viral strains expected to circulate during each hemisphere's winter season. To enable more timely antigenic assessment, we developed a deep learning framework that leverages genomic surveillance data to predict antigenic differences from hemagglutinin (HA) protein sequences. This approach integrates a protein language model trained on large-volume HA sequence datasets with transformer encoders for antigenic prediction. Compared with existing computational methods, our approach demonstrates improved performance in both cross-validation and forward seasonal nowcasting tasks. We further developed model interpretation methods that can identify critical HA residues associated with major antigenic transitions. Applying this framework to H3N2 and H1N1 strains up to 2022, our sequence-based approach is able to predict HAI titres not measured through antigenic surveillance. By augmenting HAI titre data with sequence-derived predictions, our method enhances antigenic mapping and refines representations of viral population structure. These findings underscore the potential of integrating genomic surveillance with interpretable deep learning to enable real-time prediction of influenza antigenic drift and to advance our understanding of viral evolutionary trajectories.