药动学（PK）研究是评价药物成药性、安全性与临床适用性的关键环节，传统实验方法耗时长、成本高，人工智能（AI）的发展为解决这一困境提供了新的思路。系统梳理AI预测小分子药物PK参数的最新研究进展。在算法层面，介绍随机森林、支持向量机、梯度提升树、极端梯度提升等传统机器学习算法，以及图神经网络、深度神经网络等深度学习算法的优势、局限性及适用场景，并提出不同数据规模下的算法选择建议；在应用层面，归纳上述算法在预测药物吸收、分布、代谢、排泄等环节关键PK参数中的应用进展，以及国内外药企开展AI预测PK参数产业化研究的现状；在平台层面，总结ADMET Lab 3.0、Deep-PK、admetSAR 3.0、ADMET-AI、SwissADME等代表性在线预测平台的特点，并将各平台的半衰期预测结果与临床实测数据进行对比分析。旨在为AI预测和PK领域的研究工作者提供实践参考，助力加速药物早期筛选流程，降低研发成本与风险。

Pharmacokinetics (PK) research is a crucial step in evaluating the pharmacological properties, safety, and clinical applicability of drugs. The traditional trial-and-error experimental method is time-consuming and costly. The development of artificial intelligence (AI) provides a new solution to this problem. This paper systematically reviews the latest research progress in AI for predicting the pharmacokinetic parameters of small molecule drugs. At the algorithm level, it introduces traditional machine learning algorithms such as random forest, support vector machine, gradient boosting tree, and extreme gradient boosting, as well as deep learning algorithms such as graph neural networks and deep neural networks, their advantages, limitations, and applicable scenarios, and proposes algorithm selection suggestions for different data scales; At the application level, it summarizes the application progress of these algorithms in predicting key PK parameters such as drug absorption, distribution, metabolism, and excretion, as well as the current situation of industrialized research on AI predicting PK parameters carried out by domestic pharmaceutical companies; At the platform level, it summarizes the characteristics of representative online prediction platforms such as ADMET Lab 3.0, Deep-PK, admetSAR 3.0, ADMET-AI, and SwissADME, and conducts a comparative analysis of the half-life prediction results of each platform with clinical measured data. The aim is to provide practical references for researchers in the fields of AI prediction and pharmacokinetics, and to help accelerate the early drug screening process and reduce research and development costs and risks.

R969.1

国家自然科学基金资助项目（82607746）；天津市科技计划项目（24ZXZSSS00480）