Bayesian integrated testing strategy to assess skin sensitization potency: from theory to practice

Area of Science:

• Toxicology
• Computational Chemistry
• Dermatology

Background:

• Current toxicological frameworks struggle to integrate diverse data (in vitro, in silico, in chemico) for predicting in vivo effects.
• Quantitative, mechanistic approaches are rare, with 'weight of evidence' being primarily qualitative.
• An Integrated Testing Strategy (ITS) using a Bayesian Network (BN) was previously developed as a proof of concept.

Purpose of the Study:

• To present an updated ITS methodology for assessing skin sensitization potency.
• To incorporate enhanced mechanistic information into the Bayesian Network structure.
• To demonstrate the practical utility of the updated ITS for regulatory and research applications.

Main Methods:

• Development of an updated Bayesian Network (BN) structure for the Integrated Testing Strategy (ITS).
• Parameterization of the BN ITS using an extended dataset of 124 chemicals.
• Validation analysis and case study application to assess prediction accuracy for hazard and Local Lymph Node Assay (LLNA) potency classes.

Main Results:

• The updated BN ITS achieved high prediction accuracy: 95% for hazard and 86% for LLNA potency classes on a test set of 21 chemicals.
• The ITS demonstrated the ability to generate hypotheses from minimal data subsets (as few as one data point).
• The framework allows querying the value of additional tests before commencement, optimizing resource allocation.

Conclusions:

• The updated BN ITS provides a robust and mechanistically interpretable framework for skin sensitization assessment.
• The ITS offers significant practical value beyond mere prediction, aiding in strategic testing design.
• This approach represents a significant step towards mechanistic interpretation in toxicological testing strategies.