摘要
药物不良反应(ADR)是全球药物警戒关注的首要问题,个体遗传差异,尤其是药物基因组学(PGx)特征,是导致ADR发生的关键因素。近年来,人工智能(AI)技术为整合多组学数据、精准预测ADR提供了可能。本文系统梳理了基于PGx预测ADR的AI方法。首先整理了常用的PGx与ADR相关的多源异构数据集,然后重点列举了传统机器学习(如支持向量机、随机森林等)及深度学习(如卷积神经网络、图神经网络等)等AI模型在该领域的应用实例。这些模型通过挖掘基因变异、临床用药特征与ADR之间的复杂非线性关系,实现了ADR的智能预测。然而,该领域仍面临数据异质性、模型可解释性及临床转化障碍等挑战。文章最后展望了多模态数据融合、可解释AI等未来研究方向,旨在推动个体化安全用药和精准医疗事业的发展。
Adverse drug reaction(ADR)represents a primary concern in global pharmacovigilance.Individual genetic variations,particularly pharmacogenomics(PGx)characteristics,are key factors contributing to the occurrence of ADR.In recent years,artificial intelligence(AI)technologies have enabled the integration of multi-omics data for accurate ADR prediction.This review summarizes AI methods for predicting ADR based on PGx.It begins by organizing commonly used multi-source heterogeneous datasets related to PGx and ADR,then highlights application examples of AI models-such as traditional machine learning(e.g.,support vector machine,random forests)and deep learning(e.g.,convolutional neural networks,graph neural networks)-in this field.These models enable intelligent prediction of ADR by uncovering complex non-linear relationships among genetic variations,clinical medication features,and ADR.However,the field still faces challenges,including data heterogeneity,model interpretability,and obstacles in clinical translation.Finally,the review outlines future research directions,such as multi-modal data fusion and explainable AI,aiming to advance the development of personalized medication safety and precision medicine.
作者
韩芳芳
刘静欣
柴克燕
吴嘉瑞
蔡永铭
HAN Fangfang;LIU Jingxin;CHAI Keyan;WU Jiarui;CAI Yongming(School of Medical Information and Engineering,Guangdong Pharmaceutical University,Guangzhou 510006,China;School of Chinese Materia Medica,Beijing University of Chinese Medicine,Beijing 102488,China)
出处
《药物流行病学杂志》
2026年第1期104-113,共10页
Chinese Journal of Pharmacoepidemiology
基金
广东省中医药管理局中医药科研项目(20251212)
广东省医学科学技术基金(C2025084)
广州市科技计划项目(2025A03J3712)。
关键词
药物不良反应
药物基因组学
人工智能
机器学习
多源异构
神经网络
Adverse drug reaction
Parmacogenomics
Artificial intelligence
Machine learning
Multi-source heterogeneous
Neural network
