AI-driven prediction of monoclonal antibody binding and activity against influenza A virus

Monoclonal antibodies (mAbs) are a cornerstone of modern therapeutics for diseases ranging from cancer to infectious pathogens, yet their development remains constrained by time-intensive screening and high costs. We present an AI-driven approach to predict antibody-antigen interactions in silico, accelerating candidate selection. Focusing on influenza A virus—a persistent global threat responsible for 20–40 million U.S. infections annually—we fine-tuned a multimodal molecular language model on antibody-antigen activity data to predict binding and receptor-blocking activity against influenza A hemagglutinin (HA). The model was rigorously evaluated under varied data-split conditions to simulate real-world challenges. Our results show that even with limited fine-tuning data, this approach effectively reduces experimental burden and prioritizes high-potential mAb candidates against influenza A virus.