Decoding DNA damage: artificial intelligence and the future of genotoxicity assessment

Genotoxicity assessment is a cornerstone of chemical safety, yet conventional bioassays are often constrained by high costs, ethical concerns, and suboptimal human risk predictivity. The integration of computational toxicology and artificial intelligence (AI) has catalyzed a paradigm shift, offering scalable, data-driven solutions that enhance predictive accuracy and mechanistic insight. This review provides a comprehensive analysis of the computational frameworks driving modern genotoxicity assessment. We critically evaluate essential algorithmic approaches in machine learning (ML), deep learning (DL), and natural language processing (NLP), encompassing diverse data modalities from chemical descriptors and multi-omics datasets to high-content bioassay imagery. Key computational applications are detailed, including AI-enhanced quantitative structure-activity relationship (QSAR) modeling, automated computer vision-based screening, and NLP-driven literature mining. The review also examines the critical role of data infrastructure, including public toxicology databases, and addresses persistent challenges regarding data quality and curation. The path forward emphasizes rigorous model validation, the implementation of explainable AI (XAI) to foster regulatory trust, and integration with systems toxicology. Realizing AI's full potential to define the next generation of high-accuracy, human-relevant genotoxicity assessment is paramount and depends on robust interdisciplinary collaboration.