Advances in artificial intelligence (AI)-based tools for medicines safety evaluation, particularly in relation to Absorption, Distribution, Metabolism, Excretion and Toxicity (ADMET), offer enhanced prediction of medicine safety and performance. Building on existing in vitro and in silico approaches, AI methods can integrate diverse data sources spanning preclinical, clinical and real-world data (RWD), enabling the modelling of complex biological and clinical interactions that are not readily captured by traditional approaches.
The incorporation of RWD is a particularly important development, providing insight into how medicines behave in routine clinical practice across heterogeneous patient populations. This can support improved understanding of variability across populations, disease states and clinical conditions, more refined patient stratification, and enhanced prediction of off-target effects, including adverse drug reactions that may not be identified in preclinical studies or controlled clinical trials. AI-based approaches may also enable investigation of longer-term or otherwise difficult-to-observe outcomes and help address limitations of existing models.
Despite this potential, AI-based ADMET tools remain at an early stage of regulatory readiness. Uncertainties remain regarding model performance, evidential requirements, data quality, accessibility and the role of cross-sector data sharing in supporting robust model development and validation.
To reduce these uncertainties, a detailed exploration of current technological capabilities, data availability and regulatory considerations is required. The proposed "Beyond ADMET: AI for Medicines Safety" sandbox will take the form of iterative workshops, providing a structured regulatory “safe space” where developers can test, demonstrate and refine AI-based predictive tools intended for use in drug development and regulatory submissions. The sandbox will help build foundational knowledge, identify evidence gaps and potential regulatory needs, and establish the principles required to assess the regulatory acceptability of AI methods in medicines safety-related decision-making, supporting future pathways towards regulatory implementation.