Advancing Comet Assay Sensitivity and Specificity in Genotoxicity Research: Leveraging Artificial Intelligence and Machine Learning for <i>In Vitro</i> Toxicology

Introduction: Genotoxicity testing is critical for evaluating the safety of chemicals, pharmaceuticals, and environmental pollutants. The comet assay, or Single Cell Gel Electrophoresis (SCGE), is a widely employed method for detecting DNA damage at the single-cell level due to its sensitivity and simplicity. However, conventional manual scoring is labor-intensive, prone to observer bias, and limits the assay’s reliability and throughput. This study investigates the application of artificial intelligence (AI) and machine learning (ML) to enhance the comet assay's sensitivity, specificity, and efficiency. Methods: HepG2 cells were treated with genotoxic agents, cisplatin and doxorubicin, to induce DNA damage, followed by comet assay analysis with epifluorescence microscopy. Three ML models Support Vector Machine (SVM), Random Forest (RF), and Convolutional Neural Networks (CNN) were employed to classify comet images based on DNA damage severity. Results and Discussion: Among these, the CNN model demonstrated superior performance, achieving 92.5% accuracy and the highest correlation ( r = 0.94) with expert annotations. The AI model also quantified key parameters, including tail length, tail moment, and DNA content in the tail, offering enhanced sensitivity and specificity over manual scoring. Cross-validation and external testing validated the robustness and generalizability of the CNN model across diverse datasets. Furthermore, the AI-driven approach facilitated high-throughput analysis and minimized inter-observer variability, addressing longstanding challenges in comet assay evaluation. These results underscore the transformative potential of AI in genotoxicity testing, providing a scalable and reliable framework for advancing in vitro toxicology.