Integrated decision strategies for skin sensitization hazard

Area of Science:

• Toxicology
• Computational Chemistry
• Regulatory Science

Background:

• Skin sensitization testing is a priority for non-animal alternatives.
• The adverse outcome pathway (AOP) for skin sensitization is well-characterized.
• Integrated decision strategies using diverse data are being developed.

Purpose of the Study:

• To develop and evaluate computational models for predicting skin sensitization.
• To integrate in vitro, in chemico, and in silico data for improved prediction accuracy.
• To assess the performance of machine learning models against traditional non-animal tests.

Main Methods:

• Compiled data for 120 substances from LLNA, DPRA, h-CLAT, and KeratinoSens assays.
• Collected physicochemical properties influencing skin penetration.
• Utilized OECD QSAR Toolbox for read-across predictions.
• Built 54 models using machine learning, including support vector machines (SVM).

Main Results:

• SVM models achieved high accuracy (96-99% training, 89-96% test set) in predicting LLNA outcomes.
• SVM model performance surpassed individual non-animal tests and simple test batteries.
• Computational approaches effectively integrated diverse data sources.

Conclusions:

• Computational models, particularly SVM, show promise for non-animal skin sensitization testing.
• Integrated data strategies enhance prediction accuracy for identifying skin sensitizers.
• These methods offer a viable alternative to animal testing for skin sensitization assessment.