Ó³»­´«Ã½

Kexin Huang
Stanford Computer Science

Meeting: A General-Purpose Biomedical AI Agent

Biomedical research drives advances in human health, drug discovery, and clinical care, yet it is increasingly hindered by fragmented workflows across complex experiments, massive datasets, and vast literature. We introduce Biomni, a general-purpose biomedical AI agent that autonomously executes diverse research tasks. Biomni first employs an action discovery agent to map the biomedical action space, mining tools, databases, and protocols from tens of thousands of publications across 25 domains. Built on this foundation, its generalist architecture integrates LLM reasoning with retrieval-augmented planning and code execution, enabling dynamic composition of complex workflows without predefined templates. Benchmarking shows strong generalization across tasks such as gene prioritization, drug repurposing, rare disease diagnosis, microbiome analysis, and molecular cloning, all without task-specific tuning. Case studies further demonstrate Biomni’s ability to analyze multi-modal data and generate experimentally testable protocols. Biomni envisions AI biologists working alongside humans to accelerate biomedical discovery, clinical insight, and healthcare.

Hanchen Wang
Stanford AI Lab, Genentech

Primer: AI Agents for Biomedical Discovery: Specialized Builds and General Methods

In this talk, I’ll share our explorations on building AI Agents for biomedical discovery, focusing on perturbation design, spatial transcriptomics, and cell imaging. Beyond the scientific aspects, I’ll also discuss the technical depth of the agents we’ve built—ranging from tool-use agents and context engineering to computer use, calibration guarantees, and reinforcement learning.

Learn more about MIA.