Update and optimization of an adverse outcome pathway network of chemical-induced cholestasis

Background and objectives:
Our group previously introduced an adverse outcome pathway (AOP) network mechanistically describing key events (KEs) and their relationships driving chemical-induced cholestatic liver injury. Since AOP networks are considered to be livingdocuments and more data becomes available, they should be regularly updated. The aim of the present work was to update and optimize this AOP network in line with guidelines issued by the Organization for Economic Co-operation and Development (OECD) using new computational techniques, in particular machine learning-assisted data extraction and network analysis, in order to increase the efficiency of weight-of-evidence assessment and data visualization of the AOP network.
Methods:
PubMed was queried for studies of chemical-induced cholestatic liver injury using predefined key words and several known KE-related terms. SysRev, a recently introduced artificial intelligence (AI) tool, was used to extract data from scientific
papers in view of AOP network optimization. The tailored Bradford-Hill criteria, described by the OECD guidelines, were used for weight-of-evidence assessment of the optimized AOP network.
Results:
A labeling strategy for weight-of-evidence assessment was applied prior data extraction. A system was developed to score the extracted data compliant with OECD guidelines based on 3 categories.
Conclusions:
AI-based tools combined with the use of a newly established scoring strategy, relying on the tailored Bradford-Hill criteria, resulted in an optimized AOP network of chemical-induced cholestatic liver injury. The optimized AOP network will be used as the mechanistic backbone for the establishment of an in vitro test battery, with focus on monitoring transporter changes, to predict cholestatic liver injury induced by chemicals.