中国药物警戒 ›› 2025, Vol. 22 ›› Issue (8): 869-875.
DOI: 10.19803/j.1672-8629.20250236

• 药源性心脏毒性预警研究专栏（一） • 上一篇    下一篇

中草药心脏毒性预测模型研究

陈思颖1, 丁雪丽1, 刘淑佳1, 张晓朦1,2, 张冰1,2, 林志健1*   

  1. 1北京中医药大学中药学院,北京 102488;
    2北京中医药大学中药药物警戒与合理用药研究中心,北京 102488
  • 收稿日期:2025-04-15 出版日期:2025-08-15 发布日期:2025-08-13
  • 通讯作者: *林志健,男,博士,教授,临床中药的安全性与有效性研究。E-mail：linzhijian83@126. com
  • 作者简介:陈思颖,女,硕士,中药的安全性与代谢疾病防治研究。
  • 基金资助:
    国家自然科学基金资助项目（82274117）; 国家中医药管理局高水平重点学科建设项目-临床中药学（zyyzdxk-2023257）

Modeling for Prediction of Cardiotoxicity of Chinese Herbal Medicines

CHEN Siying1, DING Xueli1, LIU Shujia1, ZHANG Xiaomeng1,2, ZHANG Bing1,2, LIN Zhijian1*   

  1. 1School of Chinese Materia Medica, Beijing University of Chinese Medicine, Beijing 102488, China;
    2Center for Pharmacovigilance and Rational Use of Chinese Medicine, Beijing University of Chinese Medicine, Beijing 102488, China
  • Received:2025-04-15 Online:2025-08-15 Published:2025-08-13

PDF (PC)

21

摘要： 目的 利用机器学习方法,基于定量构效关系建立并验证中草药心脏毒性预测模型,为中草药的安全性评价和临床合理用药提供参考。方法 利用比例失衡法筛选并收集美国食品药品监督管理局（Food and Drug Administration, FDA）不良事件报告系统（Adverse Event Reporting System, FAERS）数据库中具有潜在心脏风险的活性成分,作为主分析数据,随机划分为训练集和验证集,分别通过随机森林（RF）、决策树（DT）、K-最近邻分类法（KNN）和极端梯度提升法（XGBoost）构建预测模型及内部验证,使用曲线下面积（AUC）、准确率、精确度等多个指标评估模型的性能,选出最优模型;并通过检索相关数据库建库至2025年1月1日的文献收集具有心脏毒性的中草药活性成分,从自发报告系统数据库挖掘出可能具有心脏风险的中草药,并从TCMSP数据库中筛查其活性成分,作为测试集,对构建的最优模型进行外部验证。结果 预测性能最佳的模型为KNN,训练集AUC=0.684,验证集AUC=0.718;通过文献筛选出24种具有心脏毒性的中草药活性成分,自发报告系统数据库中筛查出怀疑中草药11种,经外部验证后18种中草药活性成分以及10种中草药预测出心脏风险。结论 模型预测的整体准确率达80%,可以用于中草药活性成分的心脏毒性预测。

关键词: 中草药, 心脏毒性, 机器学习, 构效关系, 瓜拉纳, 预测模型

Abstract: Objective To establish a prediction model for cardiotoxicity of Chinese herbal medicines (CHMs) in order to provide data for safety evaluation and rational clinical use of CHMs. Methods Active ingredients with potential cardiac risks were identified from the FDA Adverse Event Reporting System (FAERS) database by using the proportional imbalance method. The data was randomly divided into a training set and a validation set. The Random Forest (RF), Decision Tree (DT), K-Nearest Neighbors(KNN), and Extreme Gradient Boosting (XGBoost) were used for modeling and internal verification. The performance of the model was evaluated using such criteria as the area under the curve (AUC), accuracy, and precision. Active ingredients of CHMs with cardiotoxicity were retrieved from literature that was published from the inception to January 1, 2025. CHMs with possible cardiac risks were retrieved from the spontaneous reporting system database. The Traditional Chinese Medicine Systems Pharmacology (TCMSP) database was searched for the active ingredients. The constructed model was externally validated via these tests. Results The model with the best predictability was the KNN. The AUC was 0.684 for the training set and 0.718 for the validation set. Twenty-five chemical ingredients of CHMs with cardiotoxicity were selected based on literature while eleven suspected cardiotoxic CHMs were selected from the spontaneous reporting system database. After external validation, eighteen chemical ingredients and ten CHMs were predicted to have cardiac risks. Conclusion The overall prediction of the model is 80% accurate, so it can be used for predicting cardiotoxicity of chemical ingredients in CHMs.

Key words: Traditional Chinese Medicine, Cardiotoxicity, Machine Learning, Quantitative Structure-Activity Relationship, Paullina cupana Kunth, Modeling for Prediction

中图分类号: