【摘要】 以2011—2019年山东中医药大学附属医院的高血压电子病历建立数据集，用Apriori算法量化镇肝熄风汤组成中药的配伍强度，卷积神经网络量化中药的剂量特征信息，蛋白质相互作用的拓扑特征分析生物特征信息，并将上述信息输入K最邻近分类算法、支持向量机、梯度递增决策树、贝叶斯网络和逻辑回归模型，评估各个模型效力，从多个维度体现镇肝熄风汤的组成配伍规律和作用机制。结果表明，K最邻近分类算法在5个模型中结果最优（AUC=93.5%），共获得了87组有效配伍，验证了镇肝熄风汤配伍调节细胞因子、减少炎症反应和代谢紊乱的作用机制。同时，外部测试表明此研究可外推至其他疾病，表明此模型可有效融合卷积神经网络数据和网络拓扑数据，将剂量信息和生物特征融入中药配伍挖掘，对传统的关联规则模型进行补充，具有良好的普遍适用性和外推性，为多模态的中药数据集挖掘提供了一种跨学科研究方法。更多还原

【Abstract】 The high prevalence of hypertension and its wide range of complications have made it a significant risk factor for cardiovascular disease mortality. Zhengan xifeng decoction(ZXD) is a classic traditional Chinese medicine prescription used in the clinical treatment of hypertension. However, there is a lack of systematic analysis of its composition rules and biological effects. In order to further explore the clinical application characteristics of ZXD in the treatment of hypertension, this study established a dataset based on the electronic medical records of hypertension patients from the Affiliated Hospital of Shandong University of Traditional Chinese Medicine from 2011 to 2019. Apriori algorithm was used to quantify the compatibility strength of the Chinese medicines in ZXD, while convolutional neural networks were employed to quantify the dosage characteristics.Topological feature analysis of protein interactions was used to examine biological characteristics. The outcome was then input into k-nearest neighbors, support vector machines, gradient boosting decision trees, Bayesian networks, and logistic regression models to evaluate the efficacy of each model, reflecting the composition rules and mechanisms of action of ZXD from multiple dimensions. The results showed that the k-nearest neighbors algorithm performed the best among the five models(AUC=93.5%),identifying 87 effective combinations, validating the mechanisms by which ZXD regulates cytokines, reduces inflammation, and corrects metabolic disorders. External testing indicated that this research could be extrapolated to other diseases. This study demonstrates that the model effectively integrates convolutional neural network data and network topology data, incorporating dosage information and biological characteristics into the exploration of Chinese medicine combinations. It complements traditional association rule models and has good general applicability and extrapolation capability, providing an interdisciplinary research approach for mining multimodal Chinese medicine datasets.更多还原