MOOSE-Chem demonstrates how large language models can autonomously generate valid chemistry research hypotheses from minimal background information.

• Successfully rediscovers known chemistry findings without explicit training
• Generates hypotheses that align with expert evaluation criteria
• Creates novel chemical insights without requiring specialized datasets
• Establishes a framework for AI-assisted scientific discovery in chemistry

Biological Impact: This research opens pathways for accelerated drug discovery, improved understanding of biochemical processes, and potential breakthroughs at the chemistry-biology interface by automating hypothesis generation.

MOOSE-Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses