AI Applications in Predictive Toxicology Supporting Safety Assessments

The application of artificial intelligence (AI) technologies, such as machine learning models (e.g., QSAR) and expert rules-based systems, is a well-established practice in regulatory predictive toxicology, where they are used to support drug safety assessments.

The implementation of these methods in regulatory predictive toxicology is partly due to their adherence to principles defined in the OECD QSAR Assessment Framework (QAF) that outline criteria (based on earlier validation principles for models) to help ensure data quality, objective measures of QSAR model performance and robustness, a domain of applicability, documentation, transparency, and an underlying mechanistic interpretation where possible.

Recently, advancements in technology, such as algorithmic innovation, have led to a surge in new AI modalities. New guidelines need to be developed to ensure data reliability for models to be suitable to support regulatory assessments. In this webinar, the expert speakers will illustrate how Leadscope’s QSAR models and structural alerts adhere to the QAF and support regulatory safety assessments. They will identify current and emerging regulatory guidelines and demonstrate the applicability of structure-based predictive models.

Regulatory interest in the use of QSARs to support drug safety assessments began in the 1990s. The evaluation of DNA-reactive (mutagenic) impurities in pharmaceuticals, as outlined in the ICH M7 guideline published in 2014, currently represents the most prevalent application of QSARs in predictive toxicology for regulatory purposes. It is valuable to examine the factors contributing to the successful adoption of QSARs in this context. Leadscope’s bacterial mutation consensus model of the Ames assay was developed by several key steps: curating data from trusted sources, normalizing chemical structures, determining model domain, validating performance externally and ensuring transparency and interpretability of chemical descriptors and predictions, supported by mechanistic and statistical evidence. Scientific understanding is important for refining and developing regulatory models and methodologies. For example, insights into the α-hydroxylation metabolic activation pathway led to the development of the carcinogenic potency categorization approach (CPCA) in 2024, which estimates acceptable intake limits for N-nitrosamines drug substance-related impurities (NDSRIs).

In 2020, the US Food and Drug Administration (FDA) published guidance on evaluating drug-drug interactions (DDI), discussing the use of structural features to determine the need for in vitro studies to assess cytochrome P450 (CYP) inhibition by potential metabolites. Assessing DDI is crucial as CYP enzyme inhibition or induction can alter co-administered drug activity, potentially causing adverse effects and market withdrawal. The expert speakers will examine DDI mechanisms and the use of QSARs to predict reversible and irreversible CYP inhibition as well as bioactivation alerts.

The regulatory interest in structure-based machine learning methods continues to increase. Emerging guidelines such as the ICH Q3E may provide additional insight on, if or how existing SAR knowledge could be used to conduct a safety assessment for extractables and leachables. Additionally, the demonstrated fitness for the purpose of acute oral toxicity models has triggered discussions on their applicability in regulatory contexts.

Register for this webinar today to explore the application of advanced AI technologies such as QSAR models and expert rules-based systems in supporting drug safety assessments.

Speakers

Kevin P. Cross, PhD, Head of Science, Instem

Kevin P. Cross, PhD, is the Head of Science at Instem and is Principal Investigator of US FDA/Instem research collaborations. He has been developing chemoinformatics tools and products for over 40 years.

He is involved in several collaborative efforts creating in silico protocols and procedures for performing chemical hazard and risk assessments for regulatory purposes as well as developing and assessing the performance of QSAR models. He has published over 45 papers and 3 book chapters.

Message Presenter

Candice Johnson, PhD, Senior Research Scientist, Instem

Candice Johnson, PhD, is a Senior Research Scientist at Instem. Dr. Johnson has co-authored several peer-reviewed publications and book chapters describing the implementation of in silico approaches and methodologies for gaining confidence in an in silico prediction.

Her work expands into the novel application of in silico approaches and supports the advancement of alternative methods. She is particularly interested in the application of computational tools to support toxicological evaluations, e.g., in the assessment of extractables and leachables

Message Presenter