Harvard Medical School Team Open-Sources AutoScientists, Decentralized AI Research System

According to Beating, a collaborative team from Harvard Medical School, the Kempner Institute, and the Broad Institute, including researchers Shanghua Gao, Ada Fang, and Marinka Zitnik, has open-sourced AutoScientists, a decentralized AI agent system for scientific discovery. Unlike centralized systems with single-threaded search, AutoScientists eliminates the central coordinator, enabling agents to collaborate asynchronously—agents draft peer reviews before consuming compute resources, preventing redundant failed experiments and discovering multiple promising research directions simultaneously.

In BioML-Bench testing across medical imaging, drug discovery, and protein engineering tasks, the system achieved 74.4% average leaderboard percentile across 24 tasks, improving 8.3 percentage points over prior agent baselines. On protein binding prediction, AutoScientists discovered methods that improved Spearman correlation by 6.5% on ProteinGym, surpassing previous supervised benchmarks.