Bridging Nano-Interface Interactions and Organ-Specific Toxicity: A Review of Machine Learning for Nanomaterials Risk Assessment

Risk assessment of engineered nanomaterials (ENMs) is essential for protecting human health and the environment. Traditional hazard assessments rely primarily on in vivo testing, which faces technical challenges in extrapolation validity, ethical dilemmas, and high costs. Machine learning (ML) models offer alternative approaches that are aligned with the 3R principles (Replacement, Reduction, and Refinement) for reducing animal use. ML methods help address the economic, ethical, and temporal limitations of traditional nanotoxicology while advancing mechanistic understanding. This review presents a cross-scale framework integrating nano–bio/nano–environmental interfaces, organ-specific toxicity, in vitro-to-in vivo extrapolation (IVIVE), interpretable ML, and regulatory translation. Future directions include building comprehensive databases to replace sparse literature data, developing ML models that bridge in vitro and in vivo nanotoxicity, incorporating co-exposure scenarios of nanomaterials and chemicals, and further exploring protein/lipid corona formation and structures.