Stanford researchers created a virtual biotech company made up of 37,000 artificial intelligence agents, each assigned a role in the drug-development process.
The agents analyzed and catalogued about 50,000 clinical trials in less than a week. They looked for patterns in how precisely a drug targeted a cell type and how switch-like the activity of its target gene was.
In the study, drugs aimed at more specific, switch-like targets were 48 percent more likely to reach the market and were linked with 32 percent fewer adverse events. The analysis found the pattern across conditions including cancer, brain, heart, kidney, and lung diseases.
The virtual team also proposed an antibody-drug design for lung cancer that a pharmaceutical company later developed independently. The researchers stress that the new ideas still need testing in real laboratories, where people and physical experiments remain essential.
