Alignment-free prediction of cross-reactivity in influenza A (H3N2) anticipates antigenic drift.

FluEmbed predicts influenza A (H3N2) antigenic drift with high accuracy, achieving a Spearman correlation of 0.67-0.80 from RNA sequences without alignments.

• Why it matters: Accurate prediction of antigenic evolution is crucial for timely vaccine updates, yet existing methods rely on alignments and phylogenetic models that are time-consuming and less adaptable to rapid viral changes.
• What they did: The study developed FluEmbed, a protein language model-based framework that quantifies antigenic impact directly from RNA sequences, outperforming traditional sequence-distance and phylogenetic approaches, and conducted in-silico mutagenesis to explore mutation effects.
• The result: FluEmbed enables fast, alignment-free antigenic predictions, revealing mutation patterns associated with immune escape and intra-clade competition, thereby supporting real-time influenza surveillance and vaccine design.