TCM-Agent: Advancing Network Pharmacology and Herbal Medicine Discovery with LLM-Based Multi-Agent Systems

Network pharmacology has emerged as a pivotal approach for deciphering the complex “multi-component, multi-target” mechanisms underlying traditional Chinese medicine (TCM). However, despite extensive research efforts, a comprehensive and intelligent automated analytical framework remains elusive. Large language model (LLM)-based intelligent agent systems demonstrate robust capabilities in semantic understanding, logical inference, and task orchestration. In this study, we present the first LLM-powered multi-agent system specifically designed for network pharmacology and herbal medicine research, namely TCM-Agent. The system demonstrates core capabilities including autonomous knowledge reasoning, data analysis, interactive visualization, as well as literature retrieval and validation. Benchmark evaluations across 100 validated TCM studies demonstrated that the TCM-Agent demonstrated competitive performance in answer accuracy, literature retrieval precision, and computational efficiency. Crucially, the TCM-Agent system exhibited robust and high performance across evaluated foundation model platforms (DeepSeek-v3, Qwen-plus, and GLM-4-plus). Furthermore, no significant differences were observed across the various foundation model platforms, indicating the system’s adaptability and stability when integrated with different LLM. These findings establish TCM-Agent as a robust system that provides an advanced framework, facilitating standardization, intelligent transformation, and evidence-based methodologies in network pharmacology and herbal medicine research. Consequently, TCM-Agent enhances the intelligent analysis of TCM formulas, aids in bioactive compound discovery, and establishes foundational infrastructure for next-generation network pharmacology, thereby advancing research in the field. • We present TCM-Agent, the first LLM-powered multi-agent system for autonomous network pharmacology and herbal medicine analysis, integrating knowledge reasoning, interactive visualization, and evidence-based literature validation. • By enabling multi-agent collaboration, TCM-Agent intelligently overcomes key bottlenecks, from heterogeneous data integration and analytical bias to manual verification, ultimately delivering standardized and evidence-supported interpretations. • Benchmarking across 100 validated studies demonstrates that TCM-Agent significantly outperforms existing methods in answer accuracy, literature retrieval precision, and computational efficiency. • TCM-Agent achieves closely aligned, high-performance scores across multiple foundation models (DeepSeek-V3, Qwen-Plus, GLM-4-Plus, Gemini-2.5, etc). This consistency ensures that user analysis remains robust and unaffected by the choice of a specific LLM. • Leveraging an AI-agent architecture, TCM-Agent drastically reduces analysis time from hours to minutes for network pharmacology and herbal medicine tasks, while preserving high-quality results.