Human data on remains sparse and of varying quality and reproducibility. Ex vivo experiments and animal experiments currently is the most preferred way to predict the skin sensitization approved by the regulatory agencies across the world. However, there is a constant need and demand to reduce animal experiments and provide the scope of alternative methods to animal testing. In this study, we have compared the predictive performance of the published computational tools such as ProTox-II, SuperCYPsPred with the data obtained from ex-vivo experiments. From the results of the retrospective analysis, it can be observed that the computational predictions are in agreement with the experimental results. The computational models used here are generative models based on molecular structures and machine learning algorithms and can be applied also for the prediction of skin sensitization. Besides prediction of the toxicity endpoints, the models can also provide deeper insights into the molecular mechanisms and adverse outcome pathways (AOPs) associated with the chemicals used in cosmetic products.
Keywords: computational toxicology; in silico prediction; integrated approaches to testing and assessment (IATAs); machine learning models; skin sensitization.