Development of quantitative model of a local lymph node assay for evaluating skin sensitization potency applying...

Area of Science:

• Toxicology
• Computational Chemistry
• Dermatology

Background:

• The murine local lymph node assay (LLNA) quantifies skin sensitization, but animal testing is increasingly restricted.
• Developing reliable alternative methods is crucial for cosmetic ingredient safety assessment.

Purpose of the Study:

• To develop and validate a machine learning model for predicting the LLNA EC3 value, a key indicator of skin sensitization.
• To create a non-animal testing approach for assessing chemical skin sensitization potential.

Main Methods:

• A CatBoost regression model was trained using the Cosmetics Europe database (119 substances).
• Input variables included in chemico/in vitro tests, physicochemical properties, and chemical information linked to skin sensitization pathways.
• Model performance was evaluated using the coefficient of determination (R²).

Main Results:

• The developed model achieved a high performance with R² = 0.75.
• The most influential variable identified was related to dendritic cell activation (human cell line activation test).
• The model demonstrated strong interpretability, highlighting key predictive factors.

Conclusions:

• The machine learning model provides a robust and interpretable method for quantitative skin sensitization assessment.
• This approach can serve as a valuable alternative to animal testing for regulatory purposes.
• The findings support the use of integrated testing strategies combining in vitro data and computational modeling.