Towards Bayesian-based quantitative adverse outcome pathways using in vitro data from open literature and continuous variables: a case example for liver fibrosis.

Abstract

As toxicology shifts towards non-animal testing, quantitative models are essential to predict adverse health effects from molecular or cellular perturbations. Quantitative Adverse Outcome Pathways (qAOPs) represent such models, building on mechanistic knowledge and quantifying the Key Event Relationships (KERs) described in AOPs. Despite the recognized need, the number of qAOPs remains limited. Bayesian-based approaches are often chosen for developing qAOP for their flexibility, but most use discretized variables, limiting their predictive power. In addition, these models are mainly built from newly generated data, underexploiting the large amount of information available. This study successfully leverages data from public literature and presents an innovative framework based on continuous variables to develop a Bayesian-based quantitative model for a central KER towards liver fibrosis. The model predicts the probability of the expression fold change for two key markers of hepatic stellate cell activation (aSMA and COL1A1), given the effects on tissue injury, using in vitro data from 9 chemicals. We propose a newly developed workflow to assist in knowledge identification, organization, and extraction from scientific literature and chemical databases. Based on in vitro data and in vivo information from the Open TG-GATEs (Toxicogenomics Project-Genomics Assisted Toxicity Evaluation System) database, we estimate a biologically relevant range in COL1A1 fold change that indicates an activated state of stellate cells and high liver fibrosis odds ratios. Our study provides a case example of integrating published data and continuous variables to build a Bayesian-based model, which constitutes an essential step for predicting liver fibrosis from in vitro data.