Computational toxicology studies on the interactions between environmental contaminants and biomacromolecules

Through manufacturing operations, product consumption, and drug administration, humans and wildlife are continuously exposed to environmental contaminants (ECs) throughout their lives. Faced with the potentially harmful effects of ECs on humans, regulatory agencies around the world require the integration of epidemiological, in vivo toxicological, and in vitro mechanistic data to provide the necessary information for hazard classification, labeling, and risk management of chemicals. However, animal studies have time-consuming and high-cost defects and ethical problems. High-throughput in vitro assays are also unable to provide systematic toxicological information for chemical hazard classification for over 100000 chemicals in commerce. Recently, computing resources and artificial intelligence have innovatively improved the accuracy and speed of machine learning (ML) algorithms, and the structural biology and deep learning (DL) algorithms (e.g., AlphaFold2 and AF2Complex) have incrementally resolved a large number of biomolecular crystal structures. Thus, the use of computational toxicology techniques in environmental toxicology has increased significantly. It is estimated that computational toxicology techniques can perform virtual screening of millions of compounds in a limited amount of time for the contaminant-biomolecule interaction process. Thus, computational toxicology techniques can reduce the initial experimental cost of identifying environmental emerging contaminants, increase information on the toxicity mechanisms of ECs, and improve the efficiency of hazard identification of ECs by regulatory authorities. This study systematically reviews computational toxicology techniques commonly used to analyze ECs-biomolecule interactions, including molecular docking, molecular dynamics (MD) simulation, and machine learning-based modelling. Molecular docking is a well-established molecular simulation method that explores the interactions between biomolecules and small molecules to predict their binding modes and binding affinities. MD can simulate the flexible binding process of contaminant-biomolecule and the dynamic conformational shift process of contaminant-biomolecule complexes, providing more comprehensive information on the interaction mechanism. Machine learning-based modelling is a novel computational toxicology technique that is completely different from molecular simulation. It mainly uses publicly available structural information and in chemico, in vitro, and in vivo bioactivity data to construct (quantitative) structure-activity relationships (Q)SARs based on ML algorithms, and uses (Q)SAR models to rapidly improve the efficiency of virtual screening of ECs targeted at biomolecules, and further deepen the analysis of contaminant-biomolecule interaction mechanisms in complex biological contexts. In this paper, the main applications of these techniques in the field of environmental toxicology in recent years are systematically reviewed, including the mechanistic studies of contaminant-biomolecule interactions and high-throughput virtual screening. The advantages and limitations of these techniques in terms of ligand-receptor interaction, explainability, training efficiency, computational depth, and biological processes are discussed. Results showed that MD simulations, which can deeply explore contaminant-biomolecule interactions, are unable to achieve the high-throughput virtual screening of ECs. Machine learning-based modelling, which can reflect the complex biological processes and achieve high accuracy prediction, unable to effectively interpret the prediction results due to the 'black box' defects. Molecular docking, which has the potential to be a high throughput virtual screening method, is limited by local sampling and approximate scoring function deficiencies. These problems limit the ability of molecular docking to analyze more comprehensive mechanisms of ECs-biomolecule interactions. Therefore, only the combination of multiple computational toxicology techniques to develop an integrated workflow for mechanistic analysis and virtual screening can compensate for their respective shortcomings and obtain optimal results. However, how to increase the computational throughput screening while maintaining the mechanism analysis remains a key issue to be addressed in the future.