Artificial intelligence (AI) and advanced protein language models are revolutionizing the design of monoclonal antibodies that can prevent or mitigate life-threatening viral infections, according to a new multi-institutional study led by Vanderbilt University Medical Center (VUMC).
Pioneering Research in Therapeutic Antibody Design
The study, published in the journal Cell, highlights how AI can streamline the development of antibody-based therapeutics targeting viruses such as RSV (respiratory syncytial virus) and avian influenza. However, the researchers emphasized that the implications of their work extend far beyond infectious diseases.
“This study is an important early milestone toward our ultimate goal—using computers to efficiently and effectively design novel biologics from scratch and translate them into the clinic,” said Dr. Ivelin Georgiev, the paper’s corresponding author, and Professor of Pathology, Microbiology and Immunology at Vanderbilt. He also directs the Vanderbilt Program in Computational Microbiology and Immunology.
Broad Applications Across Disease Areas
Dr. Georgiev noted that such AI-driven methods could have a profound impact on public health. “These approaches can be applied not only to infectious diseases but also to cancer, autoimmune disorders, neurological conditions, and many others,” he explained.
Georgiev, a recognized leader in computational disease modeling, collaborated closely with Dr. Perry Wasdin, a data scientist from his lab and the study’s first author, along with a team of scientists from the U.S., Australia, and Sweden.
Protein Language Model Generates Functional Antibodies
As reported by medicalxpress, the researchers demonstrated that a protein language model can design functional human antibodies capable of identifying the unique antigen sequences—or surface proteins—of specific viruses. Remarkably, the model achieved this without relying on any existing antibody sequences as templates.
Protein language models, similar to large language models (LLMs) that power AI chatbots like ChatGPT, are trained on vast datasets of protein sequences. These models can then predict new sequences with desired structural and functional properties.
MAGE: The Monoclonal Antibody Generator
In this study, the team developed MAGE (Monoclonal Antibody Generator), a protein language model trained on previously characterized antibodies against a known strain of H5N1 avian influenza. MAGE successfully generated new antibodies effective against a related but previously unseen influenza strain.
According to the researchers, these results demonstrate that MAGE “can generate antibodies against an emerging health threat far more rapidly than traditional antibody discovery methods,” which often depend on patient blood samples or viral antigens from the novel pathogen.
Collaboration and Future Impact
Alongside Drs. Georgiev and Wasdin, other Vanderbilt co-authors include Dr. Alexis Janke, Dr. Toma Marinov, Gwen Jordaan, Olivia Powers, Dr. Matthew Vukovich, Dr. Clinton Holt, and Alexandra Abu-Shmais.
This breakthrough marks a significant leap forward in AI-driven biologic design, setting the stage for faster, more adaptive responses to infectious diseases and beyond.