Integrated skin sensitization assessment based on OECD methods (III): Adding human data to the assessment.

Area of Science:

• Toxicology
• Computational Chemistry
• Risk Assessment

Background:

• New Approach Methodologies (NAMs) are essential for skin sensitizer potency assessment.
• Existing regression models for predicting a point of departure (PoD) were trained on Local Lymph Node Assay (LLNA) data.
• The Reference Chemical Potency List (RCPL) integrates LLNA and human data but is limited by sample size.

Purpose of the Study:

• To integrate diverse data sources for improved skin sensitizer potency prediction.
• To retrain regression models using an expanded dataset of potency values (PV).
• To compare the predictivity of models trained on LLNA, PV, and human DSA04 data.

Main Methods:

• Development of an enlarged database of PV (n=139) with associated in vitro data.
• Retraining regression models using LLNA, PV, and human DSA04 datasets.
• Comparative analysis of model predictivity and parameter weighting.

Main Results:

• Predictive models trained on PV achieved similar predictivity to LLNA-based models.
• PV-trained models showed reduced emphasis on cytotoxicity and increased focus on cell activation and reactivity.
• The human DSA04 dataset was too small and biased for reliable potency prediction.

Conclusions:

• An enlarged set of PV values is a valuable tool for training predictive models.
• PV-based models offer a complementary approach to LLNA-only databases for risk assessment.
• Future efforts should focus on expanding the PV dataset for robust model development.