Researchers use artificial intelligence to create functional antibodies against deadly viruses, without needing a physical sample (Figure 1).
Figure 1: Graphical abstract.
A new study published in Cell demonstrates how artificial intelligence (AI) can accelerate the design of life-saving antibody therapies. The research shows that “protein language models”, a specialized form of AI, can learn the rules of protein structure and function well enough to generate new monoclonal antibodies capable of recognizing and neutralizing viruses such as RSV and avian influenza.
Large language models (LLMs) like ChatGPT learn from vast amounts of text to predict and generate words. Similarly, protein language models are trained on sequences of amino acids, the “words” that make up proteins, allowing them to predict and design entirely new biological molecules.
Using this approach, the team developed MAGE (Monoclonal Antibody Generator), a protein language model trained on thousands of antibody sequences. Remarkably, MAGE could design functional human antibodies that recognized viral surface proteins without using any existing antibody templates as a starting point.
In one experiment, MAGE was trained on antibodies targeting a known strain of the H5N1 avian influenza virus. The AI then generated new antibodies that successfully bound to a related but previously unseen H5N1 strain, demonstrating its potential to respond to emerging viral threats faster than traditional antibody discovery methods, which typically require patient samples or purified viral proteins.
The implications extend far beyond infectious disease. AI-powered design could revolutionize treatments for cancer, autoimmune, and neurological disorders, enabling scientists to rapidly prototype custom biologics tailored to specific disease targets.
By understanding the “grammar” of protein sequences, models like MAGE open a path toward on-demand antibody design, a future where therapeutic molecules can be computationally generated within days of a new outbreak. Such technologies could one day allow us to respond to emerging health threats in real time, fundamentally transforming how medicine is developed and delivered.
AI is learning to “speak protein” and that linguistic leap could redefine how we create antibody drugs, turning computers into rapid-response tools against infectious diseases and beyond.
Journal article: Wasdin, P. T., et al., 2025. Generation of antigen-specific paired-chain antibodies using large language models. Cell.
Summary by Stefan Botha
