Structure-aware deep learning predicts influenza antigenicity and guides vaccine strain recommendation

Influenza A viruses continuously accumulate mutations that drive antigenic drift, requiring regular updates to vaccine strains. Existing sequence-based prediction methods often miss the structural changes that determine antigenicity. Vir3D leverages ESMFold, an AI system for protein structure prediction, to generate structural information directly from amino acid sequences, enabling more precise antigenic characterization. This structure-aware approach could enhance global influenza surveillance and help prioritize candidate vaccine viruses for seasonal and pandemic preparedness. The model represents a step toward integrating structural biology with machine learning for public health applications.