Increasing Accessibility of Bayesian Network-Based Defined Approaches for Skin Sensitisation Potency Assessment

Area of Science:

• Toxicology
• Computational Chemistry
• Dermatology

Background:

• Skin sensitisation is a critical safety concern for chemicals.
• Defined Approaches (DAs) integrate various methods for reliable potency assessment.
• New Approach Methodologies (NAMs) offer flexibility and improved accuracy.

Purpose of the Study:

• To develop and validate two new DA models for skin sensitisation potency assessment.
• To integrate the kDPRA assay and open-source QSAR models into DA frameworks.
• To build DAs using freely available software for increased accessibility.

Main Methods:

• Development of two Bayesian network (BN)-based DA models.
• Incorporation of kDPRA (keratinoSens-based peptide reactivity assay) data for in vitro assessment.
• Inclusion of in silico inputs from open-source Quantitative Structure-Activity Relationship (QSAR) models.
• Validation against four Local Lymph Node Assay (LLNA) potency classes.

Main Results:

• The new DA models achieved prediction accuracies of 63% and 68% for LLNA potency classes.
• Performance was comparable to or better than existing BN DA models.
• Bayesian network confidence indications effectively differentiated reliable (up to 87% accuracy) from unreliable predictions.

Conclusions:

• The developed DA models offer a reliable and reproducible method for skin sensitisation potency assessment.
• Integration of kDPRA and open-source QSAR enhances DA flexibility and accuracy.
• BN confidence metrics are valuable for interpreting prediction reliability in chemical safety assessments.