The development of a computational model for predicting sensitization potential in cosmetics.

Area of Science:

• Cosmetic Science
• Toxicology
• Computational Chemistry

Background:

• Allergenic ingredient analysis is vital for cosmetic quality and safety.
• Existing prediction methods require improvement in accuracy and flexibility.

Purpose of the Study:

• To develop an improved machine learning model for predicting cosmetic ingredient sensitization.
• To increase prediction accuracy by integrating diverse data sources and ensemble modeling.

Main Methods:

• Constructed an expanded dataset of 2,862 cosmetic ingredients from human, animal, and non-animal sources.
• Standardized sensitization outcomes to binary labels following OECD TG 406 criteria.
• Employed LazyPredict for model screening and an ensemble method for model integration.

Main Results:

• Achieved AUC scores between 0.86 and 0.91 for sensitization prediction.
• Demonstrated a statistically superior AUC performance compared to previous models (Mann-Whitney U test).
• Reduced prediction error by approximately 10% compared to models using only the original LLfNA dataset.

Conclusions:

• The ensemble model integrating multiple data sources and benchmark models enhances prediction accuracy.
• This approach provides a robust foundation for cosmetic safety assessment.
• Further external validation is recommended for the developed model.