Advances in artificial intelligence and machine learning for toxicity prediction in computational toxicology: a comprehensive review

人工智能 机器学习 计算机科学 深度学习 人工神经网络 卷积神经网络 透明度(行为) 支持向量机 人工智能应用 更安全的 优先次序 生成语法 图形 风险评估 可解释性 随机森林 计算模型 人类健康
作者
Gerald Enos Shija
出处
期刊:Toxicology Mechanisms and Methods [Taylor & Francis]
卷期号:36 (9): 1549-1567 被引量:1
标识
DOI:10.1080/15376516.2026.2712550
摘要

Over the past three decades, artificial intelligence (AI) and machine learning (ML) have revolutionized computational toxicology, providing powerful tools for predicting chemical toxicity and supporting safer assessments for human health and the environment. This review offers a critical 30-year synthesis (1995-2025) that distinguishes itself from narrower prior works through its interdisciplinary integration of historical evolution, multi-omics data fusion, nanotoxicity challenges, regulatory frameworks, multi-stakeholder perspectives, and emerging hybrid and generative models. Key findings reveal a clear progression from early artificial neural networks capturing non-linear patterns in the 1990s to modern deep learning architectures such as convolutional neural networks and graph neural networks that have achieved over 85% accuracy, primarily in retrospective benchmarks on ToxCast and Tox21 datasets for endpoints including hepatotoxicity, cardiotoxicity, and nanotoxicity. However, prospective validation on novel compounds remains limited, representing a critical translational gap. Traditional machine learning methods (random forests and support vector machines) effectively handle imbalanced high-throughput screening data, facilitating multi-omics integration and applications across pharmaceuticals, pesticides, cosmetics, and nanoparticles. These approaches strengthen read-across strategies, Integrated Approaches to Testing and Assessment (IATA), and Threshold of Toxicological Concern (TTC) frameworks. Regulatory acceptance, guided by OECD principles, increasingly emphasizes explainable AI to ensure transparency and validation. In conclusion, AI/ML approaches can substantially reduce animal testing, accelerate safety evaluations, and address data gaps, yet require ongoing attention to biases, model opacity, and limited prospective performance. Hybrid mechanistic-AI models, federated learning, and strengthened cross-sector collaboration represent the most promising path forward.

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
1秒前
616完成签到,获得积分20
1秒前
1秒前
秋风的应助被简简单单采纳,获得10
2秒前
2秒前
越来越没意思了完成签到,获得积分10
2秒前
茉莉花完成签到,获得积分10
2秒前
3秒前
4秒前
秋风的应助被Panda采纳,获得10
4秒前
5秒前
5秒前
changdingN发布了新的文献求助10
6秒前
yff完成签到,获得积分10
6秒前
7秒前
syvshc的应助被sweeeeeet采纳,获得10
7秒前
8秒前
8秒前
SciGPT的应助被pp采纳,获得10
8秒前
coasting发布了新的文献求助10
8秒前
Echo给Echo的求助进行了留言
9秒前
秋风的应助被机智凌雪采纳,获得20
9秒前
9秒前
传奇3的应助被梦想的笨猪采纳,获得10
9秒前
AN完成签到 ,获得积分10
9秒前
风趣怀亦完成签到 ,获得积分10
9秒前
9秒前
沙丁鱼的猫完成签到 ,获得积分10
10秒前
10秒前
April发布了新的文献求助10
10秒前
酷波er的应助被太空工程师采纳,获得10
11秒前
华仔的应助被周一采纳,获得10
11秒前
MOJITO完成签到,获得积分20
12秒前
12秒前
WYJ发布了新的文献求助10
13秒前
小马甲的应助被陈月亮采纳,获得50
14秒前
14秒前
少川完成签到 ,获得积分10
15秒前
16秒前
16秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 2000
Handbook of Sport Coaching Research 1000
Orthognathic Surgery - 2 Volume Set 500
高分子基功能界面水蒸发材料与应用研究 500
外场增强的水汽发电研究 500
基于结构调控的功能化水诱导发电材料与器件研究 500
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8193690
求助须知:如何正确求助?哪些是违规求助? 9579864
关于积分的说明 21868057
捐赠科研通 7793415
什么是DOI,文献DOI怎么找? 3437414
关于科研通互助平台的介绍 2560832
邀请新用户注册赠送积分活动 2484206