August 9, 2025
5 min read
PYMNTS
Stanford AI Agents’ Lab Accelerates Discovery of Promising COVID-19 Nanobody Drug Leads
Stanford researchers have developed a virtual laboratory powered by artificial intelligence (AI) agents that rapidly designed molecules targeting fast-evolving COVID-19 variants. This innovative approach enabled the creation of 92 novel nanobody candidates in just days, two of which demonstrated strong binding to strains that evade existing antibody therapies. Professor James Zou of Stanford University, who led the project, envisioned AI agents modeled after his lab to explore research ideas beyond the limited time human scientists have. The AI agents sifted through trillions of molecular combinations to identify promising candidates, a task that would have taken humans months to accomplish.Novel Use of Nanobodies
Unlike traditional antibodies commonly used in drug discovery, the AI agents focused on nanobodies—tiny, rare cousins of antibodies typically found in animals like camels. Nanobodies are smaller, easier to computationally model, and potentially more stable, making them attractive therapeutic candidates. "If we asked most human researchers to design binders, many would probably have said, ‘Let’s try antibodies,’" Zou explained. "Nanobodies are much less common, so it’s an interesting and surprising decision by the virtual lab." Physical experiments confirmed that the 92 AI-designed nanobodies bound effectively to the latest coronavirus strains, outperforming existing human-designed antibodies.AI Agents Collaborate Like a Research Team
The virtual lab is powered by GPT-4o from OpenAI and features two lead AI agents: the Principal Investigator (PI) and the Scientific Critic (SC). The PI manages a team of specialized AI agents, including an immunologist, computational biologist, and machine learning specialist. These agents collaborate through team and individual meetings, debating strategies such as whether to design nanobodies or antibodies. The SC rigorously critiques the agents’ proposals to minimize errors and hallucinations, ensuring robust conclusions. Multiple parallel meetings are conducted to synthesize consensus responses, enhancing reliability. The PI agent summarizes discussions and decisions for human researchers, who oversee the project at a high level but contribute only about 1% of the total work. In this project, out of 122,462 words generated, the human researcher wrote just 1,596.Impact and Future Applications
The AI-driven virtual lab discovered two novel COVID-19 vaccine candidates more effective against recent variants and better at protecting against the original strain. The platform has been receiving enthusiastic feedback and is being adapted to other challenges, such as identifying biomarkers for Alzheimer’s disease. Importantly, the virtual lab is fully open source and available on GitHub, allowing researchers worldwide to download, modify, and apply it to diverse scientific problems. Zou’s findings are detailed in a paper published in Nature, co-authored with John Pak of the Chan Zuckerberg Biohub and Stanford computer science graduate Kyle Swanson.AI’s Growing Role in Healthcare
The use of AI in drug discovery is gaining momentum. Google DeepMind’s CEO Demis Hassabis recently predicted AI-designed drugs will enter clinical trials by the end of 2025, signaling a new era in pharmaceutical innovation.Conclusion
Stanford’s AI agents demonstrate that collaboration between human scientists and AI can dramatically accelerate research, tackling complex problems like COVID-19 drug discovery more efficiently than either could alone.“Collaboration between AI and human is certainly much more effective than either the human alone or the AI alone,” said Zou. “In problems like medicine and diseases, there’s actually no shortage of problems to solve.”
Source: Originally published at PYMNTS on August 8, 2025.