The system, designed by Stanford researchers, identified which drugs are more likely to succeed in trials and even proposed a cancer treatment a major drugmaker later landed on too.
Image Credit
Myriam Zilles on Unsplash
Share
Developing a new drug can take years and cost hundreds of millions of dollars, and even then, most candidates ultimately fail. Now, researchers at Stanford have built a virtual biotech company with 37,000 AI agents that work together to analyze drug targets and design therapies.
Roughly 90 percent of drugs that enter clinical trials never reach the market. That’s often because promising results in the lab don’t translate to patients, or the drug causes dangerous side-effects not caught earlier in the development process.
Part of the problem is the evidence that could help catch these issues earlier in the process is scattered across disciplines and formats, making it hard for any single team to weigh it all.
To get around this, a Stanford team created a system they call a virtual biotech, which consists of up to 37,000 AI agents built to mimic the divisions of a real drug-development company. In a paper published in Science, the system identified which types of drug targets are more likely to succeed in clinical trials and even proposed a lung cancer treatment that a major drugmaker later landed on too.
“Our idea was to see how far we could push this. Could we create a biotech company that takes on everything from looking for drug targets all the way to designing clinical trials?” senior author James Zou said in a press release.
The new system features a virtual chief scientific officer (CSO) that takes a query from a human user and then delegates tasks to an army of specialized “scientist” agents working on the problem.
These agents are armed with their own databases and tools and are split into one of four divisions that specialize in finding and validating drug targets, assessing safety risks, choosing how a drug should be delivered, and reviewing existing clinical trial data. The system has built-in access to the Open Targets database, a massive public repository of clinical trial data.
To test the system, the researchers gave it an existing study showing that genetic evidence can help predict which drugs succeed in trials and asked it how to build on that research. The CSO decided the first step was to improve the quality of the data it had access to because many trials in the Open Targets database don’t clearly record whether the drug actually worked.
So, it asked its researcher agents to dig through the outcomes of 37,075 individual Phase II and III trials, assigning one agent to each trial. The agents searched trial registries, published papers, and press releases for results. They crunched through the job in about six hours—a fraction of the time it would take a team of humans.
The CSO asked another agent to look for promising gene candidates by scouring a public database of human tissues showing which genes are switched on in which cell types. It came up with a two-part scoring system, which first measured whether a gene was active in just one type of cell or across many and then gauged whether its activity was controlled more like an on-off switch or could be dialed up and down like a dimmer switch.
Be Part of the Future
Sign up to receive top stories about groundbreaking technologies and visionary thinkers from SingularityHub.
Comparing those scores to the updated trial outcome data revealed a pattern. Drugs aimed at switch-like genes only found in a small number of cell types were 48 percent more likely to eventually reach the market, 40 percent more likely to advance from Phase 1 to Phase 2 trials, and had 32 percent fewer adverse events than drugs hitting more broadly active targets.
The researchers then pushed the system further, asking it to evaluate a protein called B7-H3 that’s associated with lung cancer. The agents discovered the protein was particularly common in connective-tissue cells called fibroblasts that are often found close to tumor cells.
The agents then discovered evidence those cells were suppressing the activity of nearby immune cells, preventing the body from detecting and reacting to the tumors. The system proposed a therapy that would tag cells expressing B7-H3 with an antibody to help direct a toxic chemotherapy drug to them.
The virtual biotech came up with its solution based solely on data available before January 2025, but in August of that year a major pharmaceutical company arrived at the same strategy independently, when its B7-H3-targeted therapy ifinatamab deruxtecan received FDA breakthrough therapy status. “This was really exciting as an independent, third-party validation that’s consistent with the effects and the design proposed by the virtual biotech,” Zou said.
However, coming up with drug targets is just one step in a long, expensive drug discovery process. While refining the candidate selection process could prevent drug companies from pursuing some obvious dead ends, it can’t speed up the rigorous lab testing and clinical trials required to get a drug to market.
Nonetheless, given the industry’s woeful record at translating promising science into finished products, an army of AI scientists that can significantly speed up a critical part of the drug discovery pipeline could be just what the doctor ordered.
Related Articles
The Next Frontier Is Not Artificial Intelligence—It’s Artificial Societies
Single CAR T Injection Eases Multiple Sclerosis Symptoms in Small Trial
Google’s Genome Atlas Predicts the Effect of Every Possible DNA Mutation
What we’re reading
Facts Only
* Stanford researchers built a virtual biotech company with 37,000 AI agents.
* The system analyzed drug targets and designed therapies.
* The system identified which drug targets are more likely to succeed in clinical trials.
* The system proposed a lung cancer treatment based on the analysis.
* The system featured a virtual chief scientific officer (CSO) that delegated tasks.
* Agents were split into four divisions: finding/validating targets, assessing safety risks, choosing drug delivery, and reviewing trial data.
* The system accessed the Open Targets database of clinical trial data.
* Agents analyzed 37,075 Phase II and III trials to review outcomes.
* Agents searched for promising gene candidates in human tissue databases.
* The system developed a two-part scoring system for gene activity (on-off switch vs. dimmer switch).
* Drugs targeting switch-like genes found in few cell types were 48% more likely to reach the market.
* The system proposed a therapy targeting B7-H3 in fibroblasts associated with tumors.
* A major pharmaceutical company independently pursued a B7-H3-targeted therapy.
Executive Summary
A virtual biotech system, developed by Stanford researchers, utilizes 37,000 AI agents organized to mimic the divisions of a real drug-development company to analyze drug targets and design therapies. The system aimed to address the high failure rate in drug development, where most candidates do not reach the market, often due to issues missed during early stages. The system incorporates a virtual chief scientific officer who delegates tasks to specialized agent teams focused on finding drug targets, assessing safety risks, choosing delivery methods, and reviewing clinical trial data.
The researchers tested the system by analyzing 37,075 Phase II and III trials from the Open Targets database to identify patterns in successful drug targets. The agents analyzed trial outcomes and gene activity, leading to a scoring system that favored genes associated with "switch-like" activity found in limited cell types as more likely to succeed clinically. This analysis led to proposing a specific treatment strategy for lung cancer targeting B7-H3 and its association with tumor microenvironments.
The results suggested that drugs targeting switch-like genes in specific cell types had a higher probability of clinical success and fewer adverse events. This predictive modeling was validated when the proposed B7-H3 therapy strategy was independently pursued by a major pharmaceutical company, achieving breakthrough status for an analogous treatment.
Full Take
The creation of the virtual biotech illustrates a significant shift from reactive scientific inquiry to predictive, parallel processing in R&D. The core implication is that complex systems involving biological discovery, which are currently bottlenecked by human cognitive capacity and data fragmentation, can be accelerated by specialized, interconnected agents operating on massive datasets. The system’s success hinges not just on the volume of data (37,000 agents reviewing thousands of trials) but on the intelligent delegation architecture—the CSO framework that allows for hierarchical reasoning about quality control before generating proposals.
The pattern observed suggests that identifying specific mechanistic signatures (e.g., gene switching behavior in specific cellular contexts) provides predictive power that transcends traditional linear hypothesis testing. The fact that an AI-derived strategy was independently validated by industry entities lends weight to the methodology, suggesting the system successfully navigated the high dimensionality of drug discovery into a tangible, reproducible intervention.
The risk lies in assuming that this acceleration translates perfectly to novel biological uncertainties. If the initial training data (Open Targets) contains inherent biases regarding trial outcomes or safety reporting, the AI might optimize for known limitations rather than discover fundamentally new pathways. The focus should shift from simply accelerating the *selection* of targets to developing mechanisms for the agents to autonomously challenge established scientific paradigms, ensuring that the emergent solutions are not merely optimized versions of existing, potentially flawed, biological assumptions.
Bridge Questions: What governance structures are necessary to ensure the safety and ethical oversight of autonomous systems proposing novel therapies? How can researchers design feedback loops within the virtual system to intentionally explore counter-intuitive or underrepresented data points rather than confirming known patterns? What institutional changes are required to integrate these predictive capabilities effectively into the existing, slow-moving clinical validation process?