Virtual Biotech: 37,000 AI Agents Identify Potential Lung Cancer Drug Targets

Imagine a pharmaceutical company with tens of thousands of employees working around the clock to identify the next successful drug. An advanced artificial intelligence system called Virtual Biotech is taking a step toward that vision.

Described in the journal Science¹, Virtual Biotech consists of 37,000 AI agents that can interact autonomously with large language models (LLMs) and complete complex, multistep tasks.

The system identified molecular signals that may help predict whether clinical trials will succeed. With human oversight, it also identified a potential lung cancer treatment targeting the protein CD276.

“We’re looking forward to seeing how far an agent team of AI scientists can actually help accelerate drug discovery and development,” said James Zou, a computer scientist at Stanford University in California who led the study.

However, other scientists caution that virtual biotechnology systems have not yet been tested in the real-world challenges of drug discovery. Their predictions have also not been confirmed through experiments or clinical trials.

How the Virtual Biotech AI System Works

AI systems are increasingly being applied to complex biomedical tasks, including genomic data analysis, scientific hypothesis generation and experimental design.

To assess whether AI agents could help discover new medicines, Zou assembled a team designed to resemble the structure of a biotechnology company. A chief scientific officer, or CSO, agent directs “employees” assigned to specialized departments such as drug-target identification and clinical-trial design.

According to the study, Zou’s team used a version of Claude, developed by Anthropic in San Francisco, California, as the underlying LLM. Zou said that other advanced LLMs could also be used, including open-source models that researchers can run on their own computers.

AI Agents Analyze More Than 55,000 Clinical Trials

To test Virtual Biotech, the researchers analyzed published results from more than 55,000 clinical trials involving drugs for a range of conditions. The CSO assigned 37,075 agents to examine individual late-stage trials.

Other virtual biotech agents searched datasets showing which genes are active in different cell types. Their analysis found that drugs targeting proteins active in specific cell types were nearly 50% more likely to reach the market than other drugs.

AI Identifies CD276 as a Potential Lung Cancer Target

In another demonstration, Zou’s team asked the CSO to investigate whether CD276 could be a useful therapeutic target for lung cancer. Previous research suggested that CD276 suppresses immune responses and is highly expressed in lung tumors.

Using previously collected data, the system confirmed CD276 as a potential target and developed a treatment strategy involving an antibody that recognizes CD276 and is linked to an anticancer drug.

With help from external reviewers, Zou and his collaborators concluded that the approach was promising. Nevertheless, the proposed treatment still requires experimental validation and clinical testing before its effectiveness and safety can be established.