ONTOX Insights #3

Exposure assessment is essential in environmental health research for understanding chemical risks. A systematic scoping review, provided on the Sysrev platform (web platform introducing machine learning techniques into the review process with aim for increasing accuracy and efficiency), examined human exposure assessment methods and computational tools of risk assessment from papers published since 2000, where exposure methods were properly described. This review highlighted a diverse range of chemicals, including pesticides, metals, and volatile organic compounds, and showed a significant rise in probabilistic and computational methods for exposure assessment over time. A total of 63 models and toolboxes were identified worldwide, with 12 commonly used models associated with specific exposure routes, chemical classes, and input parameters. The findings offer valuable guidance for selecting suitable methods in environmental risk assessments, particularly for projects like ONTOX. Input parameters used in each mathematical model and toolbox shown by this analysis can possibly contribute to the harmonising process across the models and tools enhancing consistency in regulatory processes and enabling better comparisons across studies.

Physiologically based kinetic (PBK) modelling offers a way to predict compound pharmaco-/toxicokinetics, helping integrate toxicity and exposure data to reduce animal testing. Traditional PBK models rely on animal and human data, limiting their use in non-animal methods. To overcome this, high-throughput PBK modelling uses only in vitro and in silico data. We evaluated various in silico tools and parameterization strategies to create PBK models, collecting over 2,000 human in vivo concentration–time profiles for 200+ compounds. Testing these models in PK-Sim, we found that high-throughput PBK modelling achieved accurate pharmacokinetic predictions for most compounds (87% of Cmax and 84% of AUC within tenfold). However, prediction accuracy varied by parameterization strategy and compound. We propose a high-throughput PBK modelling approach using only free tools, advancing confidence in PBK modelling for Next-Generation Risk Assessment.

Toxicologists constructed Adverse Outcome Pathways (AOPs) to understand how compounds cause adverse effects by mapping biological interactions that induce toxic response. Building AOPs is really intensive, relying on manual searching, collection and synthesis of scientific literature reviews. Natural Language Processing (NLP), can transform this process by systematically and quickly extracting relevant data, improving accuracy and reproducibility. We applied deep learning models to identify key toxicological entities and their causal links, focusing on two most frequent toxicities in the liver: cholestasis and steatosis. The NLP approach demonstrated effective screening of compounds and mechanisms from molecular to organismal levels. This work provides 1) proof that NLP can support modern toxicology via extraction of relevant information and 2) a template open-source model for identifying toxicological entities and their relationships.

Serum transaminases, alkaline phosphatase, and bilirubin levels are commonly used to diagnose, classify, and predict outcomes in drug-induced liver injury (DILI). However, the roles of clinical assessments, histopathology, and the chemical properties of drugs have not been fully explored. Given the complexity and prevalence of cholestasis as a DILI manifestation, our objective was to assess the importance of clinical features and drug properties in categorising drug-induced cholestasis (DIC) patients, and to develop a prognostic model to identify high-risk patients and drugs of particular concern.

Adverse Outcome Pathways (AOPs) were developed in toxicology to map out the biological events that lead to toxic effects, enhancing chemical safety assessments. AOPs help identify knowledge gaps, support therapeutic target discovery, and guide the development of in vitro and in silico test methods for hazard and risk assessment. However, because many toxicological processes are highly complex, single linear AOPs often cannot capture all relevant interactions. To address this, AOP networks have been introduced to represent interconnected toxicological events, especially as complex exposure scenarios and interactions that may influence the ultimate adverse outcome. This study constructed an AOP network based on established criteria, linking thirteen AOPs related to nephrotoxicity, sourced from the AOP-Wiki, identifying several central key events (KEs), including mitochondrial dysfunction, oxidative stress, and tubular necrosis, which are pivotal in the progression of kidney-related adverse outcomes like kidney failure and chronic kidney disease. These central KEs can offer a strong foundation for developing targeted in vitro and in silico assays to replace animal-based in vivo experiments, improving predictions and assessments of chemical-induced nephrotoxicity relevant to human health.

Follow us on social media!