Researchers developed the DeepTCM1.0 multi-agent framework for analyzing traditional Chinese medicine formulas
The research team created DeepTCM1.0 — a multi-agent framework based on the DeepSeek V3.2 model that simulates 11 interdisciplinary agents — for mechanistic analysis of traditional Chinese medicine formulas, tested on the Guizhi Decoction case.
The research team introduced the DeepTCM1.0 framework, a multi-agent system built on the general-purpose large language model DeepSeek V3.2, designed for mechanistic analysis of traditional Chinese medicine (TCM) formulas. According to the authors, existing approaches — data mining and network pharmacology — cannot adequately connect classical TCM theory with modern scientific research, and directly querying general-purpose LLMs about this field is limited by insufficient adaptation to TCM theoretical frameworks and susceptibility to hallucinations during reasoning.
The framework uses a three-layer collaborative architecture and a three-round iterative quality control process that simulates collaboration among 11 interdisciplinary agents. An analysis of the Guizhi Decoction formula served as a representative test case, in which the framework combined the perspective of classical TCM theory with modern scientific research.
The performance of the framework was evaluated using double-blind, five-dimensional scoring, reliability testing using the intraclass correlation coefficient (ICC), the Mann-Whitney U test, and effect size analysis. Four independent language models served as evaluators, each conducting five rounds of repeated evaluation of five anonymized reports, yielding a total of 100 independent evaluations.
Why it matters
According to the authors, general-purpose language models fail when directly queried about the mechanisms of traditional Chinese medicine because of insufficient adaptation to its theoretical frameworks and susceptibility to hallucinations, while older analytical methods (data mining, network pharmacology) cannot connect classical theory with modern research. DeepTCM1.0 is presented as an attempt to address this gap through structured collaboration among multiple specialized agents and systematic, repeatedly validated evaluation of outputs instead of a one-off response from a single model.
Check the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.