ONTOX Insights #4

Toxicological test methods generate raw data and guidelines for processing it to classify test compounds as hits or non-hits. The data processing pipeline described within a test method is often highly complex, involving multiple layers of data, from machine-generated outputs to the final classification. Each transition between these layers requires several data processing steps, and any modifications to these steps can significantly impact the final outcome of New Approach Methodologies (NAMs). Transparent documentation of these processes is crucial to ensuring the robustness, performance, and credibility of NAMs. Reporting templates like ToxTemp should include detailed descriptions of data processing to maintain consistency and reproducibility. The same raw data, if processed differently, may yield varying results, impacting the readiness status of a NAMs. This work provides an overview of key data levels in NAMs development and the processing steps between them. As NAMs become integral to toxicological risk assessment, raising awareness of data-handling practices is essential for fostering trust, ensuring reproducibility, and supporting regulatory acceptance.

The workshop "State of the Science on Assessing Developmental Neurotoxicity Using New Approach Methods" was co-organized by JIFSAN and the FDA’s Human Foods Program and hosted by the FDA in College Park, MD, on 14-15 November 2023. It brought together experts from international organizations, government agencies, industry, and academia to discuss the shift from traditional in vivo tests to innovative New Approach Methods (NAMs) for Developmental Neurotoxicity (DNT) testing. The discussions emphasized the vulnerability of the developing brain to toxic exposures and the need for ethical, cost-effective, and scientifically robust alternatives. Experts explored various NAMs, including in silico, in vitro, and non-mammalian models, while addressing challenges in replicating human neurodevelopment and integrating DNT NAMs into regulatory frameworks. The workshop provided a comprehensive overview of current DNT NAMs, assessed their strengths and limitations, and outlined key steps for advancing regulatory acceptance of these innovative methods.

The integration of AI into new approach methods (NAMs) is transforming chemical safety assessment, offering opportunities to streamline validation and enhance predictive capabilities. This review examines key challenges, such as data quality, model interpretability, and regulatory acceptance, while highlighting AI’s potential to improve predictive power and data integration. The concept of e-validation is introduced as an AI-driven framework to modernize NAMs validation through automated reference chemical selection, study simulation, mechanistic validation, and model training and evaluation. Tiered approaches, performance benchmarking, uncertainty quantification, and cross-validation are proposed as validation strategies. The importance of continuous monitoring and human oversight is emphasized, addressing ethical considerations in AI-driven toxicology. Looking ahead, the review explores emerging trends in AI development, research priorities, and the role of collaboration among scientists, regulators, and industry to integrate AI-based NAMs into toxicological practice. The authors describe the vision of companion AI post-validation agents to keep methods and their validation status up to date. By addressing these challenges, AI can enhance predictive toxicology while reducing animal testing and increasing human relevance and translational capabilities.

Toxicology has evolved from an observational science into a data-rich discipline, creating new opportunities for AI integration. The expansion of computing power and the accumulation of large toxicological datasets have created the way for applying machine learning and deep learning techniques to enhance chemical hazard assessment. This article explores the key developments in AI-enabled toxicology, tracing its evolution from early expert systems and statistical learning methods, such as quantitative structure-activity relationships (QSARs), latest deep neural network approaches and emerging trends. It examines both the promises and challenges of AI adoption in predictive toxicology, data analysis, risk assessment, and mechanistic research. While AI holds significant potential for accelerating evidence-based toxicology and improving human health and environmental protection, it is not a universal solution. Interpretable and human-centered AI tools, developed through multidisciplinary collaboration, are essential to ensure responsible application. AI should be thoughtfully designed and integrated alongside ongoing efforts to enhance primary evidence generation and appraisal, fostering a balanced approach to modern toxicological research.

The EU REACH (Registration, Evaluation, Authorisation and Restriction of Chemicals) Regulation requires animal testing only as a last resort, yet the authors study from 2023 (Knight et al., 2023) reveals that by December 2022, around 2.9 million animals had already been used for REACH testing on reproductive, developmental, and repeated-dose toxicity. An estimated 1.3 million more will be required for ongoing tests, with additional testing expected as compliance checks continue. With 75% of read-across methods rejected by the European Chemicals Agency (ECHA), reliance on animal testing remains high. Authors estimated that 0.6 to 3.2 million animals have been used for other toxicological endpoints, and discussions on grouping 4,500 registered petrochemicals could further affect these numbers. The 2022 REACH amendment is projected to add 3.6 to 7.0 million more test animals. Meanwhile, upcoming legislative proposals could push these numbers even higher, including extending Chemical Safety Assessment (CSA) to Annex VII substances, potentially adding another 1.6 to 2.6 million animals, and the registration of polymers, presenting challenges similar to the petrochemical debate.

Follow us on social media!