摘要
Traditional Chinese Medicine (TCM) faces a persistent “black box”problem arising from multicomponent, multi-target, and nonlinear synergistic mechanisms that challenge reductionist biomedical paradigms. This review synthesizes major breakthroughs (2024 2025) and argues that computational ethnopharmacology represents a paradigm shift beyond classical network pharmacology.We propose a four-layer architecture—data, semantic, topological, and structural layers—integrating evidence-oriented databases (e.g., HERB 2.0), GraphRAG-enabled LLM mining of classicaltexts, hypergraph/heterogeneous GNN modeling of formula compatibility, and AlphaFold 3 drivenstructural inference coupled with inverse docking and molecular dynamics. Focusing on clinicallyactionable herb drug interaction prediction, we further outline validation and reporting checkliststo improve reproducibility and safety translation, and highlight dry wet closed-loop directions forfuture research.
