Juraj Gottweis, Wei‐Hung Weng, Alexander Daryin, Tao Tu, Petar Sirkovic, Anatoly Myaskovsky, Grzegorz Glowaty, Felix Weissenberger et al.
Co-Scientist, a multi-agent AI system built on Gemini, automates and accelerates scientific discovery by generating and refining research hypotheses for experimental validation.
Scientific discovery relies on generating hypotheses for complex problems and rigorous experimental validation, a process that is time and resource-intensive. Existing AI systems lack effective support for structured scientific thinking and novel hypothesis generation.
A multi-agent architecture based on Gemini was designed with an asynchronous task execution framework for flexible test-time compute scaling. A tournament evolution process enables agents to continuously generate, critique, and refine hypotheses for self-improving hypothesis generation.
Validation was conducted in three biomedical applications: drug repurposing, novel-target discovery, and explaining antimicrobial resistance mechanisms. Specifically, it identified new drug-repurposing candidates and synergistic combination therapies for acute myeloid leukaemia, which were validated through in vitro experiments, demonstrating its potential to accelerate real-world scientific discovery.