Frameworks to predict in vivo effects by integration of in vitro, in silico and in chemico information using mechanistic insight are needed to meet the challenges of 21(st) century toxicology. Expert-based approaches that qualitatively integrate multifaceted data are practiced under the term 'weight of evidence', whereas quantitative approaches remain rare. To address this gap we previously developed a methodology to design an Integrated Testing Strategy (ITS) in the form of a Bayesian Network (BN). This study follows up on our proof of concept work and presents an updated ITS to assess skin sensitization potency expressed as local lymph node assay (LLNA) potency classes. Modifications to the ITS structure were introduced to include better mechanistic information. The parameters of the updated ITS were calculated from an extended data set of 124 chemicals. A detailed validation analysis and a case study were carried out to demonstrate the utility of the ITS for practical application. The improved BN ITS predicted correctly 95% and 86% of chemicals in a test set (n = 21) for hazard and LLNA potency classes, respectively. The practical value of using the BN ITS is far more than a prediction framework when all data are available. The BN ITS can develop a hypothesis using subsets of data as small as one data point and can be queried on the value of adding additional tests before testing is commenced. The ITS represents key steps of the skin sensitization process and a mechanistically interpretable testing strategy can be developed. These features are illustrated in the manuscript via practical examples.
Keywords: Bayesian Network; LLNA potency class; integrated testing strategy; skin sensitization.
Copyright © 2013 John Wiley & Sons, Ltd.