Integrative modeling of seasonal influenza evolution via AI-powered antigenic cartography

Seasonal influenza viruses evade host immunity through rapid antigenic evolution. Antigenicity is measured by serological assays and visualized as antigenic maps. However, conventional maps cannot infer the antigenicity of uncharacterized strains directly from sequence data. Here, we present PLANT, a protein language model that embeds influenza A/H3N2 viruses onto antigenic maps from HA sequences, enabling comprehensive antigenic maps that encompass all sequenced strains. Using PLANT, we demonstrate that (i) antigenic evolution accelerates during disrupted global circulation such as the COVID-19 pandemic, (ii) antigenic change is a major driver of viral fitness advantage, and (iii) PLANT enables automated early detection of immune-escape variants and systematic evaluation of historical vaccine strains. We further establish a PLANT-based vaccine strain selection framework that offers improved antigenic match compared with WHO recommendations. By linking viral genotype, antigenicity, and fitness, this study advances understanding of influenza evolution and provides a powerful framework for rational vaccine design.