筛查持久性、生物蓄积性有毒化学品的环境计算毒理学:进展与展望

Environmental computational toxicology for screening persistent,bio-accumulative,and toxic chemicals:Progress and perspectives

筛查有害化学物质,有助于源头防控化学品的风险,防范化学品成为新污染物.具有环境持久性(P)、生物蓄积性(B)的有毒(T)化学品,是典型的有害化学物质.高的环境持久性,容易导致化学品在环境中积累,导致其高环境暴露浓度.高生物蓄积性的化学品,容易在生物体内具有高的内暴露浓度.毒性则直接决定化学品的环境风险.本文综述了化学品P,B,T属性值的环境计算毒理学预测方法和模型,以及基于机器学习的PBT化学品的集成筛查技术,展望了机器学习算法在化学品P,B,T属性值预测、PBT化学品的集成筛查方面的应用前景.融合机器学习等技术,环境计算毒理学有望从化学品P,B,T属性值预测、有害化学品筛查、替代化学品设计的角度,助力化学品风险管理和新污染物治理.

In the process of production,storage,transport,application and disposal,chemicals may be released or discharged into the environment and become emerging pollutants.Once chemicals become dispersed contaminants in environmental media such as the atmosphere,water and soil,its removal requires additional energy,which could be expended to generate other pollution.Therefore,screening of hazardous chemicals is a prerequisite for controlling their risks at source and preventing them from becoming emerging pollutants.Persistent,bio-accumulative,and toxic(PBT)chemicals are of great concern due to their environmental risks.In particular,environmentally persistent chemicals tend to accumulate in the environment,leading to increased environmental exposure concentrations.Bio-accumulative chemicals tend to have high internal exposure concentrations in organisms.Meanwhile,toxicity,as significant as exposure characteristics,is directly related to environmental risks.This paper summarizes cutting-edge research in environmental computational toxicology for screening PBT chemicals.There are two general ideas for predicting the environmental persistence of chemicals:Mechanistic profiling and data-driven modelling.Mechanistic profiling refers to predicting the total degradation rate constant or half-reduction period of a chemical by modelling its degradation and transformation behavior in the multimedia environment.Degradation pathways of chemicals in different environmental mediums and degradation kinetics parameters of each pathway vary.Therefore,it is necessary to comprehensively analyze occurrence and distribution of chemicals in atmospheric(photodegradation and free radical oxidative degradation),water(biodegradation,photodegradation and hydrolysis),soil(biodegradation)and biological(metabolic transformation)phases,as well as the key degradation and transformation processes.It was found that machine learning-assisted molecular dynamics simulations can improve computational efficiency and facilitate revealing environmental behavior of chemical pollutants.Along with the accumulation of data on the kinetics parameters of the environmental degradation and transformation of chemicals,various machine learning algorithms can be applied to construct data-driven quantitative structure-activity relationship(QSAR)models.Prediction and simulation of chemical bioaccumulation can be achieved using data-driven approaches and toxicokinetic models based on the absorption distribution,metabolism and elimination processes.Artificial intelligence techniques such as transfer learning and multi-task learning can enable high-throughput prediction of toxic effect endpoints at different biological levels,and can potentially contribute to better model prediction performance and broader applicability domain.In vitro-in vivo extrapolation(IVIVE)methodology can qualitatively or quantitatively infer in vivo toxic dose levels from in vitro bioactivity data.IVIVE is an important risk prioritization tool of chemicals and has been validated for risk assessment of certain chemicals.Future research can integrate multimodal data such as molecular structure,omics data,high-throughput and high-content screening results to enable personalized and precise prediction of bioactivity related endpoints.Rapid developments in deep learning techniques have made it possible to reveal complex patterns behind big data.With a list of hazardous chemicals,it is possible to develop models for high-throughput screening of hazardous chemicals,given the principle that molecular properties are determined by their structures.Machine learning based comprehensive screening techniques for PBT chemicals were introduced,including graph attention network models that can identify PBT chemicals directly from molecular structures.Finally,challenges and perspectives of environmental computational toxicology techniques in PBT chemical screening were discussed.Incorporating techniques such as machine learning,environmental computational toxicology is expected to facilitate chemicals management and emerging pollutants control from the perspectives of chemical P,B,T property prediction,hazardous chemical screening and alternative chemical design.