Biomni: A General-Purpose Biomedical AI Agent.

Area of Science:

• Biomedical research and artificial intelligence.

Background:

• Biomedical research is increasingly complex, facing limitations from fragmented workflows and data overload.
• There is a critical need for advanced solutions to scale scientific expertise and accelerate discovery.

Purpose of the Study:

• Introduce Biomni, a general-purpose AI agent designed for autonomous execution of biomedical research tasks.
• Establish a unified agentic environment by mining tools, databases, and protocols from extensive biomedical literature.

Main Methods:

• Biomni utilizes an action discovery agent to map the biomedical action space.
• Employs a generalist agentic architecture integrating large language model (LLM) reasoning, retrieval-augmented planning, and code-based execution.
• Dynamically composes and executes complex biomedical workflows without predefined templates.

Main Results:

• Biomni demonstrates strong generalization across diverse tasks like gene prioritization, drug repurposing, and rare disease diagnosis.
• Achieves high performance without task-specific prompt tuning.
• Successfully interprets multi-modal biomedical data and generates experimentally testable protocols in real-world case studies.

Conclusions:

• Biomni represents a significant advancement in automating and scaling biomedical research.
• Envisions AI biologists augmenting human scientists to enhance research productivity and clinical insights.
• Offers a platform for collaborative innovation in the next era of biomedical discovery.