GoKawiilAfter examining the results of thousands of clinical trials, a swarm of AI agents suggested that a protein that dampens immune responses could be the target for a promising treatment for lung cancer.Credit: Eoneren/Getty
It’s a pharmaceutical-company executive’s dream: tens of thousands of employees working day and night, uninterrupted by meals, sleep or distraction, to find the next blockbuster drug.
A sophisticated artificial-intelligence system called the Virtual Biotech, which is described today in the journal Science1, is a step towards this vision.
The Virtual Biotech comprises as many as 37,000 agents — AI systems that autonomously interact with large language models (LLMs) or with each other and are capable of performing multistep tasks. It uncovered a molecular signal that could help to predict clinical-trial success. And with some human oversight, it identified a promising lung-cancer treatment.
"We want to see how far these agent teams of AI scientists can help us to really accelerate drug discovery and development,” says James Zou, a computer scientist at Stanford University in California who led the effort.
But other scientists note that the Virtual Biotech has not been vetted in the crucible of real-world drug discovery, and its predictions were not validated through experiments, let alone clinical trials.
Organizational chart
In the past year or so, ever-more-capable AI scientists have taken hold in various fields. In biomedicine, they have shown aptitude for complex tasks ranging from genomic data analysis to hypothesis generation and experimental design.
To test these systems’ ability to find new drugs, a job that involves numerous interrelated tasks, Zou assembled a team of agents, mirroring the staff of a biotechnology company. In the Virtual Biotech’s set-up, a chief scientific officer (CSO) agent directs ‘employees’ in different divisions, each with their own subspecialities, such as target identification and clinical-trial design.
For the Science study, Zhou’s team used versions of Claude — developed by Anthropic in San Francisco, California — as the underlying LLM that powered the agents. But he says that any advanced LLM will do, including open-source models that researchers can run on their own computers.
... continue reading