Modeling skin sensitization: hierarchical support vector regression−based prediction of lysine depletion in DPRA

Skin sensitization is a critical endpoint in toxicology, especially in the context of drug discovery and development process for topical and transdermal treatments. Accurate evaluation of skin sensitization potential is essential to ensure the safety and efficacy of new compounds before their incorporation into commercial products. In this study, a quantitative structure−activity relationship (QSAR) model employing a hierarchical support vector regression (HSVR) scheme was developed to quantitatively predict the percentage of lysine depletion measured by the Direct Peptide Reactivity Assay (DPRA) that is one of the keys factors used to identify the skin sensitization potential of chemical compounds. The results demonstrated that the HSVR model produced accurate and consistent predictions across the training, test, and outlier sets, underscoring its robustness and reliability across various datasets. Moreover, the HSVR model showed outperformed other available models in terms of predictive accuracy. These findings highlight the potential of computational modeling as a rapid, cost−effective, and ethically sound alternative to traditional methods. This advancement represents a significant contribution to safer product development in the cosmetic and pharmaceutical industries.