피부과민성 예측 향상을 위한 QSAR 모델의 비교 및 통합적 접근

Toward Improved Prediction of Skin Sensitization:Comparative and Integrated QSAR Approaches

Skin sensitization is one of the key adverse effects caused by exposure to chemicals, cosmetics, or pharmaceuticals, and is a major endpoint required in product development and risk assessment. Due to the ethical, time, and cost limitations of animal testing, quantitative structure–activity relationship (QSAR) models have emerged as promising New Approach Methodologies (NAMs). In this study, the predictive performance of QSAR models (Derek Nexus, TOPKAT, OECD QSAR Toolbox, and VEGA-CAESAR, VEGA-JRC, and VEGA-NSCTOX) was evaluated using domestic and international reference skin sensitizers to assess their applicability in chemical hazard and pharmaceutical safety evaluation. Among the models, OECD QSAR Toolbox showed the highest sensitivity (0.835), while TOPKAT and VEGA-CAESAR achieved values above 0.7, indicating suitability for screening positive substances, though false positives require consideration. Derek Nexus demonstrated the highest specificity (0.829), whereas others showed values below 0.5 except the OECD QSAR Toolbox (0.616). Based on these results, a multi-tiered predictive strategy is proposed: TOPKAT and VEGA-CAESAR as primary models, followed by Derek Nexus and OECD QSAR Toolbox as secondary models, to identify substances requiring priority evaluation.