AI驱动的化学品安全性评估：技术范式、应用进展与未来展望

AI-Driven Chemical Safety Assessment: Technical Paradigms, Application Progress and Future Prospects

摘要

人工智能（Artificial Intelligence, AI）技术的快速发展正在深刻重塑化学品安全性评估的传统模式。本文系统综述了 AI，尤其是机器学习和深度学习方法在计算毒理学领域的研究进展与应用现状。从数据基础、算法体系与模型可解释性这 3 个核心维度，梳理了 AI 驱动安全性评估技术范式的演进路径，并结合药物研发、化妆品安全性评估以及环境新污染物治理等典型应用场景，评述了相关技术从“替代实验手段”向“安全性赋能设计工具”的角色转变；在此基础上，文章进一步分析了国际监管框架对计算毒理学方法的接受与规范化进展，探讨了 AI 模型在可解释性、复杂体系评估、系统集成以及伦理与标准化等方面面临的关键挑战。总之，本文梳理了技术演进脉络，提出了智能毒理系统的概念框架，旨在为 AI 驱动化学品安全性评估技术的科学研究与监管应用提供系统参考与前瞻性思考。

Abstract

The rapid development of artificial intelligence (AI) is profoundly reshaping the traditional paradigm of chemical safety assessment. This paper systematically reviews the research progress and application status of AI, particularly machine learning and deep learning methods, in the field of computational toxicology. From three core dimensions—data foundation, algorithm system and model interpretability, the evolutionary path of the AI-driven safety assessment technical paradigm is summarized. Combined with typical application scenarios such as drug research and development, cosmetic safety evaluation, and the control of emerging environmental pollutants, the role transformation of related technologies from "alternative experimental methods" to "safety-enabled design tools" is discussed. On this basis, the paper further analyzes the acceptance and standardization progress of computational toxicology methods in international regulatory frameworks, and explores the key challenges faced by AI models in terms of interpretability, complex system assessment, system integration, ethics, and standardization. In summary, this paper sorts out the context of technological evolution and proposes a conceptual framework for intelligent toxicology systems, aiming to provide a systematic reference and forward-looking perspective for the scientific research and regulatory application of AI-driven chemical safety assessment technologies.

关键词

计算毒理学 / 安全性评估 / 新途径方法 (NAMs) / 不良结局通路 (AOP) / 监管科学

Key words

computational toxicology / safety assessment / new approach methodologies (NAMs) / adverse outcome pathway (AOP) / regulatory science

参考文献

[1] WANG Z, WALKER G W, MUIR D C G, et al. Toward a global understanding of chemical pollution: A first comprehensive analysis of national and regional chemical inventories[J].Environmental Science & Technology, 2020, 54(5):2575-2584.

[2] CORIA J. Policy monitor-The Economics of toxic substance control and the REACH directive[J].Review of Environmental Economics and Policy, 2018, 12(2):342-358.

[3] KREWSKI D, ANDERSEN M E, TYSHENKO M G, et al. Toxicity testing in the 21st century: Progress in the past decade and future perspectives[J].Archives of Toxicology, 2020, 94(1):1-58.

[4] RAIES A B, BAJIC V B. In silico toxicology: Computational methods for the prediction of chemical toxicity[J].Wiley Interdisciplinary Reviews: Computational Molecular Science, 2016, 6(2):147-172.

[5] YANG H, SUN L, LI W, et al. In silico prediction of chemical toxicity for drug design using machine learning methods and structural alerts[J].Frontiers in Chemistry, 2018, 6:30.

[6] WU L, YAN B, HAN J, et al. TOXRIC: A comprehensive database of toxicological data and benchmarks[J].Nucleic Acids Research, 2023, 51(D1):D1432-D1445.

[7] WEN Y, YU Y, SUN Q, et al. Study of construction of health risk and toxicity database of pollutants in China[J].Research of Environmental Sciences, 2022, 35(11):2608-2617.

[8] BAUER C, DANG D L T, VAN DEN BEUCKEN T, et al. Systematic analysis of hepatotoxicity: Combining literature mining and AI language models[J].Frontiers in Artificial Intelligence, 2025, 8:1561292.

[9] KLEINSTREUER N, HARTUNG T. Artificial intelligence (AI)-It's the end of the tox as we know it (and I feel fine)[J].Archives of Toxicology, 2024, 98(3):735-754.

[10] WU L, WANG F, ZHANG Y, et al. Multiscale-information-embedded universal toxicity prediction framework[J].Environmental Science & Technology, 2025, 59(41):21830-21847.

[11] CHITHRANANDA S, GRAND G, RAMSUNDAR B. ChemBERTa: Large-scale self-supervised pretraining for molecular property prediction[EB/OL]. (2020-10-23). https://arxiv.org/abs/2010.09885.

[12] ROSS J, BELGODERE B, CHENTHAMARAKSHAN V, et al. Large-scale chemical language representations capture molecular structure and properties[J].Nature Machine Intelligence, 2022, 4(12):1256-1267.

[13] HA S, BANG D, KIM S. Fate-tox: Fragment attention transformer for E(3)-equivariant multi-organ toxicity prediction[J].Journal of Cheminformatics, 2025, 17(1):74.

[14] LIN Z, CHOU W C. Machine learning and artificial intelligence in toxicological sciences[J].Toxicological Sciences, 2022, 189(1):7-19.

[15] CHENG F, LI W, ZHOU Y, et al. admetSAR: A comprehensive source and free tool for assessment of chemical ADMET properties[J].Journal of Chemical Information and Modeling, 2012, 52(11):3099-3105.

[16] YANG H, LOU C, SUN L, et al. admetSAR 2.0: Web-service for prediction and optimization of chemical ADMET properties[J].Bioinformatics, 2019, 35(6):1067-1069.

[17] GU Y, YU Z, WANG Y, et al. admetSAR3.0: A comprehensive platform for exploration, prediction and optimization of chemical ADMET properties[J].Nucleic Acids Research, 2024, 52(W1):W432-W438.

[18] LUECHTEFELD T, HARTUNG T. Navigating the AI frontier in toxicology: Trends, trust, and transformation[J].Current Environmental Health Reports, 2025, 12:51-69.

[19] GUENGERICH F P. Mechanisms of drug toxicity and relevance to pharmaceutical development[J].Drug Metabolism and Pharmacokinetics, 2011, 26(1):3-14.

[20] CARRACEDO-REBOREDO P, LIÑARES-BLANCO J, RODRÍGUEZ-FERNÁNDEZ N, et al. A review on machine learning approaches and trends in drug discovery[J].Computational and Structural Biotechnology Journal, 2021, 19:4538-4558.

[21] European Parliament Council of the EU. Regulation (EC) No 1223/2009. on cosmetic products (recast) [R/OL].https://eur-lex.europa.eu/eli/reg/2009/1223/oj/eng.

[22] OECD. Guideline No. 497: Defined approaches on skin sensitisation [S/OL].https://www.oecd.org/en/publications/guideline-no-497-defined-approaches-on-skin-sensitisation_b92879a4-en.html.

[23] 所雅琼, 罗飞亚, 张凤兰, 等. 皮肤致敏替代方法和整合策略研究进展及思考 [J].香料香精化妆品, 2021(5):96-102.

[24] TARIQ F, AHRENS L, ALYGIZAKIS N A, et al. Computational tools to facilitate early warning of new emerging risk chemicals[J].Toxics, 2024, 12(10):736.

[25] DULIO V, ALYGIZAKIS N, NG K, et al. Beyond target chemicals: Updating the NORMAN prioritisation scheme to support the EU chemicals strategy with semi-quantitative suspect/non-target screening data[J].Environmental Sciences Europe, 2024, 36:113.

[26] 施荣鑫, 张林, 曹泽星. 机器学习辅助预测潜在 A 类神经毒剂及其降解性质 [J].环境化学, 2025, 44(7):2727-2738.

[27] ZHANG F, WANG Z, PEIJNENBURG W J G M, et al. Machine learning-driven QSAR models for predicting the mixture toxicity of nanoparticles[J].Environment International, 2023, 177:108025.

[28] OECD. (Q)SAR Assessment framework: Guidance for the regulatory assessment of (Quantitative) structure activity relationship models, predictions, and results based on multiple predictions[R]. Paris: OECD Publishing,2023.

[29] ECHA. How to use read-across and (Q)SARs under REACH[R]. Helsinki: European Chemicals Agency,2022.

[30] U.S. Senate. S. 5002: FDA Modernization Act 2.0 (117 th Congress) [R/OL].https://www.congress.gov/bill/117th-congress/senate-bill/5002.

[31] 中华人民共和国生态环境部. 新化学物质环境管理登记指南 [R].北京: 生态环境部, 2020.

[32] 国家药品监督管理局. 国家药品监督管理局公告2021 年第 51 号: 化妆品安全评估技术导则（2021 年版）[EB/OL]. (2021-04-19). https://www.nmpa.gov.cn/xxgk/ggtg/hzhpggtg/jmhzhptg/20210419163037171.html.

[33] HAN P, LI X, YANG J, et al. Advancing toxicity predictions: A review on in vitro to in vivo extrapolation in next-generation risk assessment[J].Environment & Health, 2024, 2(7):499-513.

[34] HEMMERICH J, ECKER G F. In silico toxicology: From structure-activity relationships towards deep learning and adverse outcome pathways[J].Wiley Interdisciplinary Reviews: Computational Molecular Science, 2020, 10(4):e1475.

[35] SADEK M, KALLINA E, BOHNÉ T, et al. Challenges of responsible AI in practice: Scoping review and recommended actions[J].AI & Society, 2025, 40(1):199-215.

基金资助

国家重点研发计划课题(2023YFF1204904)

海关总署科研项目(2024HK185)