Abstract: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.

SepPCNET: Deeping Learning on a 3D Surface Electrostatic Potential Point Cloud for Enhanced Toxicity Classification and Its Application to Suspected Environmental Estrogens

QSAR Study on Estrogenic Activity of Structurally Diverse Compounds Using Generalized Regression Neural Network

A Ternary Classification Using Machine Learning Methods of Distinct Estrogen Receptor Activities Within A Large Collection of Environmental Chemicals

Interpretable attention-based multi-encoder transformer based QSPR model for assessing toxicity and environmental impact of chemicals

Interpretable machine learning for the identification of estrogen receptor agonists, antagonists, and binders

Linking electron ionization mass spectra of organic chemicals to toxicity endpoints through machine learning and experimentation

ED Profiler: Machine Learning Tool for Screening Potential Endocrine-Disrupting Chemicals

Advancing chronic toxicity risk assessment in freshwater ecology by molecular characterization-based machine learning

Computational Models to Predict Endocrine-Disrupting Chemical Binding with Androgen or Oestrogen Receptors.

Environmental Computational Toxicology for Screening Persistent, Bio-Accumulative, and Toxic Chemicals: Progress and Perspectives

In silico screening of estrogen-like chemicals based on different nonlinear classification models.

A Deep Learning Based Multi-Model Approach for Predicting Drug-Like Chemical Compound’s Toxicity

Developing QSAR Models with Defined Applicability Domains on PPARγ Binding Affinity Using Large Data Sets and Machine Learning Algorithms.

A deep learning based multi-model approach for predicting drug-like chemical compound's toxicity

QSAR Models Using a Large Diverse Set of Estrogens

Prediction of estrogen receptor binding for 58,000 chemicals using an integrated system of a tree-based model with structural alerts.

ToxMPNN: A Deep Learning Model for Small Molecule Toxicity Prediction.

Modeling and Insights into the Structural Characteristics of Endocrine-Disrupting Chemicals

Toxicity Remission of PAEs on Multireceptors after Molecular Modification Through a 3D-QSAR Pharmacophore Model Coupled with a Gray Interconnect Degree Method.

Deep Learning in Environmental Toxicology: Current Progress and Open Challenges

In Silico Prediction of Chemical Acute Dermal Toxicity Using Explainable Machine Learning Methods