AI Agents // Virtual lab for Human-Agent scientific research collaboration
Latest news from the AI Space // It’s sometimes hard to tell where sci-fi ends and reality begins
Science frequently benefits from teams of interdisciplinary researchers. However, most scientists don’t have access to experts from multiple fields. Fortunately, large language models (LLMs) have recently shown an impressive ability to aid researchers across diverse domains by answering scientific questions. Here, we expand the capabilities of LLMs for science by introducing the Virtual Lab, an AI-human research collaboration to perform sophisticated, interdisciplinary science research. The Virtual Lab consists of an LLM principal investigator agent guiding a team of LLM agents with different scientific backgrounds (e.g., a chemist agent, a computer scientist agent, a critic agent), with a human researcher providing high-level feedback. We design the Virtual Lab to conduct scientific research through a series of team meetings, where all the agents discuss a scientific agenda, and individual meetings, where an agent accomplishes a specific task. We demonstrate the power of the Virtual Lab by applying it to design nanobody binders to recent variants of SARS-CoV-2, which is a challenging, open-ended research problem that requires reasoning across diverse fields from biology to computer science. The Virtual Lab creates a novel computational nanobody design pipeline that incorporates ESM, AlphaFold-Multimer, and Rosetta and designs 92 new nanobodies. Experimental validation of those designs reveals a range of functional nanobodies with promising binding profiles across SARS-CoV-2 variants. In particular, two new nanobodies exhibit improved binding to the recent JN.1 or KP.3 variants of SARS-CoV-2 while maintaining strong binding to the ancestral viral spike protein, suggesting exciting candidates for further investigation. This demonstrates the ability of the Virtual Lab to rapidly make impactful, real-world scientific discovery.
Virtual Lab architecture. a, The workflow for designing agents in the Virtual Lab. Each agent is specified with four criteria: title, expertise, goal, and role. The human researcher in the Virtual Lab specifies these criteria to define the Principal Investigator (PI) agent and the Scientific Critic agent. Then, given a short description of the project by the human researcher, the PI agent automatically creates several scientist agents to work on the project by specifying their title, expertise, goal, and role, using its own prompt as an example. b, The workflow for a team meeting in the Virtual Lab. The human researcher writes an agenda for the meeting specifying the topic of discussion. The PI agent begins the meeting by providing initial thoughts and agenda questions as a guide for the remaining agents. Then, over the course of N rounds of discussion, each scientist agent provides its response, followed by a critique by the Scientific Critic agent, with the PI agent then synthesizing the discussion and asking follow-up questions. Finally, after the N rounds of discussion, the PI agent summarizes the discussion and provides an answer regarding the meeting agenda. c, The workflow for an individual meeting. The human researcher writes an agenda for the meeting specifying the topic of discussion. Then, the scientist agent tasked with the individual meeting provides a response to the agenda, which is critiqued by the Scientific Critic. In each round, the scientist agent improves its answer based on feedback from the Scientific Critic. Finally, after the N rounds, the scientist agent provides its final, improved answer.
Researchers have developed a novel approach to scientific collaboration, harnessing the power of artificial intelligence (AI) agents to design new SARS-CoV-2 nanobodies. The Virtual Lab, an AI-human research collaboration, brings together a team of large language model (LLM) agents with different scientific backgrounds to perform interdisciplinary science research. In a recent study, the Virtual Lab was applied to design new nanobodies that can bind to the spike protein of the KP.3 variant of SARS-CoV-2, a challenging and rapidly evolving target.
The Virtual Lab consists of a principal investigator (PI) agent, scientist agents, and a scientific critic agent. The PI agent guides the research project, while the scientist agents provide expertise in specific areas, such as immunology or machine learning. The scientific critic agent provides feedback and critiques the answers provided by the other agents. Through a series of team meetings and individual meetings, the Virtual Lab agents discuss research questions and develop solutions. In the case of SARS-CoV-2 nanobody design, the Virtual Lab created a computational workflow that incorporated state-of-the-art machine learning and computational biology tools, including ESM, AlphaFold-Multimer, and Rosetta.
The Virtual Lab designed 92 new nanobodies, which were experimentally validated by human researchers. The results showed that the designed nanobodies exhibited diverse binding profiles across multiple strains of SARS-CoV-2. Two of the designed nanobodies, Nb21 mutant I77V-L59E-Q87A-R37Q and Ty1 mutant V32F-G59D-N54S-F32S, showed promising binding profiles to the JN.1 RBD, a variant of SARS-CoV-2. These results demonstrate the potential of the Virtual Lab for rapidly generating promising scientific results that can streamline further experiments.
The Virtual Lab offers several benefits, including rapid discussion and decision-making, minimal human input required, and improved consistency and quality of answers through parallel meetings. Additionally, the inclusion of a scientific critic agent ensures that the research is rigorous and scientifically sound. The Virtual Lab architecture is also flexible and can be applied to a wide variety of interdisciplinary research projects.
The success of the Virtual Lab in designing SARS-CoV-2 nanobodies marks a significant shift in the way we approach scientific research. By harnessing the power of AI agents and combining them with human expertise, researchers can rapidly achieve promising results that can inform and accelerate further experiments. The Virtual Lab represents a new era of scientific collaboration, one that has the potential to revolutionize the way we tackle complex research questions.
