Building the Foundations of AI-Driven Toxicology: How to Use Fragmented Data for Mechanism-Based Human Health Risk Assessment

Environmental human health assessment requires reliable, comprehensive, and standardized toxicology data to support regulatory decision-making. Yet, existing databases remain fragmented, with narrow and imbalanced coverage of species and organs, incomplete dose-response relationships, and inconsistent validation chains, limiting their utility for risk prediction. Meanwhile, regulatory and technological shifts toward AI-based computational models highlight the urgency of building high-quality toxicology databases as the foundation of next-generation methodologies. This perspective outlines key challenges in data curation, harmonization, and accessibility and presents strategic solutions, including building confidence grading frameworks to leverage heterogeneous data sets, using novel high-throughput platforms to generate interaction data with high accuracy and efficiency, and facilitating community-based data sharing. We further emphasize the development of AI-enabled strategies to improve the organization, interoperability, and usability of toxicological data infrastructures in support of AI-driven environmental toxicology and mechanism-based human health risk assessment. These strategies include integrating knowledge networks to construct mechanism-informed AI models, applying transfer learning frameworks that bridge large-scale pretraining with small-sample fine-tuning, and leveraging knowledge graph enhancement and prompt learning to predict systematic "Exposure-Biology-Disease" interactions. These efforts can transform fragmented resources into systematic, interpretable, and predictive systems. We concluded that building high-quality toxicology databases can accelerate the transition to AI-driven toxicology, providing a foundation for more reliable risk assessment and stronger global environmental health protection.