Advances in the application of computational toxicology in cosmetic safety assessment.

With the global ban on animal testing for cosmetics, computational toxicology has emerged as a pivotal alternative for safety assessment. This review systematically analyzes technical frameworks, literature trends, multi-endpoint applications, and global regulatory developments. The study integrates core in silico methods, including quantitative structure–activity relationship (QSAR) models, expert systems and read-across, molecular simulations, physiologically based pharmacokinetic models, and machine learning, to evaluate their effectiveness in predicting key endpoints such as skin sensitization, genotoxicity, and endocrine disruption. Research shows that in silico methods can now predict metabolites and mixtures beyond single ingredients, with representative integrated models achieving ~86% accuracy in skin sensitization assessment, outperforming traditional assays (78%), and QSAR tools reaching 71.4–100% accuracy for OECD chemicals via in vitro data fusion. Global regulatory frameworks are gradually accepting defined approaches that integrate computational chemistry with in vitro data, with the EU, US, and China having formally adopted defined approaches for skin sensitization. However, challenges persist in model interpretability, stereoisomer coverage, and fragmented regulatory standards.  This review recommends prioritizing AI-driven multi-source data fusion to enhance transparency, comprehensive absorption-distribution-metabolism-excretion frameworks to resolve data gaps, and stereochemistry-aware evaluation to improve precision. Establishing standardized high-quality datasets, fostering interdisciplinary expertise, and constructing open data-sharing platforms will promote widespread application of computational toxicology, ultimately contributing to animal testing replacement.