机器学习方法应用于化学物质毒性测试研究进展

Research progress in applications of machine learning in toxicity prediction

随着高通量技术的发展和海量毒理学数据的涌现,毒理学研究步入大数据时代。如何高效整合现有的毒理学数据、阐明化学物质毒作用规律并利用规律提示新信息,实现新化学物质毒性高效预测,是毒理学研究的前沿问题之一。鉴于传统化学物质毒性测试方法高成本、低通量且难以揭示机制信息,迫切需求高通量预测模型。机器学习方法已应用于毒性测试,如监督学习模型、无监督学习模型、深度学习模型、强化学习模型和迁移学习模型,其常用的化学物质特征数据主要包括化学物质结构数据、文本数据、毒理基因组数据和图像数据。将机器学习应用于毒性测试的研究潜力巨大,且取得一定进展,但目前研究主要集中于数据处理和模型开发,尚未形成应用度广、共识性强的方法。此外,机器学习模型的预测精度不仅取决于算法,亦受到数据质量的影响,算法与数据质量的相互促进发展亦是一大挑战。总之,毒理学领域的数据处理和模型构建需要跨学科的合作和技术创新,随毒理学数据库的日趋完善,各种模型算法的不断优化,基于机器学习模型开展新化学物质毒性预测将日益高效、准确,对保障人类健康和环境安全起到重要作用。

With the emergence of high-throughput technology and massive toxicology data,toxicology research has entered the era of big data.How to efficiently integrate existingtoxicological data,clarify the toxic effects of chemicals,and use these patterns to providenew information,in order to achieve efficient prediction of the toxicity of new chemicalsubstances,is one of the cutting-edge issues in toxicology.In view of the high cost,low throughput and difficulty in revealing the mechanism information of traditional chemical toxicity testing methods,high throughput prediction models are urgently needed.Machine learning methods have been applied to toxicity testing,such as supervised learning models,unsupervised learning models,deep learning models,reinforcement learning models,and transfer learning models.Chemical characteristic data commonly used in machine learning models include chemical structure data,text data,toxicological genome data and image data.There is huge potential for applying machine learning to toxicity testing and machine learning methods have made some progress.However,current research focuses on the processing of data and development of models,which has failed to produce a widely used and accepted method.In addition,the prediction accuracy of machine learning models is not only dependent on algorithms,but also affected by data quality,and the mutual promotion and development of algorithms and data quality remains a big challenge.In short,data processing and model construction in the field of toxicology require interdisciplinary cooperation and technological innovation.With the increasing perfection of toxicology databases and the continuous optimization of various model algorithms,the toxicity prediction of new chemicals based on machine learning models will become increasingly efficient and accurate,playing an important role in ensuring human health and environmental safety.