DeepRoot: A KG-Coordinated Multi-Agent System for Therapeutic Reasoning over Historical Medical Texts

DeepRoot is a novel multi-agent LLM system designed for therapeutic reasoning using historical medical texts, integrating a verified knowledge graph to enhance the conversion of pre-ontological data into actionable drug discovery leads. It demonstrated a recovery rate of 47.6% for compound-disease treatment pairs from the Shen Nong Ben Cao Jing, significantly outperforming baseline LLMs and tool-using LLMs, which showed high rates of evidence hallucination. This system's architecture separates grounding from reasoning, offering a promising approach for practitioners aiming to systematically leverage historical medical knowledge in modern biomedical applications.