VirPLM: Antigenic prediction of influenza A/H3N2 viruses with a fine-tuned protein language model.

VirPLM accurately predicts antigenic evolution of influenza A/H3N2 viruses, outperforming existing methods with high robustness across seasons and evaluations.

• Why it matters: Effective prediction of antigenic drift is crucial for timely vaccine updates, but current assays are labor-intensive and low-throughput, limiting rapid response capabilities.
• What they did: The study adapted the ESM-2 protein language model to HA1 sequences from H3N2 viruses, creating VirPLM, a two-stage framework that enhances sequence-based antigenic prediction with high accuracy.
• The result: VirPLM not only surpasses existing approaches but also identifies key antigenic sites and improves strain coverage estimates, supporting better-informed vaccine strain selection.