IFTP-TCM：伝統的な中国医学における標的予測とメカニズム解明のための成分フリーの解釈可能なフレームワーク

Overcoming the chemical complexity of Traditional Chinese Medicines (TCMs) to elucidate their therapeutic mechanisms remains a central challenge in TCM research, necessitating transparent AI-driven hypothesis-generating tools. Traditional paradigms for studying holistic TCM mechanisms, such as network pharmacology (NP), rely on chemical composition data to construct "herb-compound-target" networks. However, the inherent chemical complexity of TCMs often introduces substantial noise into these herb-target associations, leading to high false-positive rates. Concurrently, conventional differential omics analysis offers limited mechanistic precision. Herein, we present IFTP-TCM, an interpretable AI framework for predicting direct targets of TCMs for predicting direct targets of TCMs. This chemical component-independent framework integrates weighted tissue-specific regulatory networks integrates multimodal biomedical data. Using a mathematical AI model, IFTP-TCM backpropagates TCM-induced gene-level changes to identify true functional targets at the protein level, providing a novel approach for holistic direct target prediction in TCMs. We evaluated IFTP-TCM using gene expression data from TCM perturbations, comparing its performance against baseline methods and established NP databases. Results confirmed its robust capability in identifying direct TCM targets. Validation using small-molecule compounds with annotated targets demonstrated that IFTP-TCM successfully predicted 10.6 % of known targets within the top 50 ranked predictions. Furthermore, applying IFTP-TCM to the mineral-based TCM Arsenic Trioxide (As₂O₃) successfully recapitulated its experimentally validated anticancer mechanisms. This underscores the method's distinct advantage in predicting TCM targets without requiring explicit chemical composition data. Notably, IFTP-TCM demonstrates unique value for studying non-botanical TCMs and processed TCM products (Pao Zhi), categories historically challenging to analyze due to chemical complexity and limitations of traditional component-dependent methods. In summary, our findings not only validate the robustness and generality of the IFTP-TCM framework but, more significantly, provide an innovative technological tool to advance precision medicine and modernization research for traditional herbal medicines.