Refinement of Structural Alerts for Hepatic Steatosis

Abstract

With the growing focus upon application of new approach methodologies (NAMs) within regulatory risk assessment of chemical substances, there is a clear need for human-relevant in silico models to predict toxicity associated with chemical exposure. This work describes the development of an in silico profiler for hepatic steatosis, a condition that results from the excess accumulation of fat within the liver. Toxicological data for chemicals were compiled from the literature, and expert judgments were made regarding their potential to induce steatosis. This list was used to test the 214 rules developed in a previous study. Many of these rules, originally derived from molecular initiating event (MIE)-based activation of ten nuclear receptor subtypes linked to liver injury, were found to be overly general, thereby limiting their predictive utility. These were supplemented with new fragments generated from the curated data set, which, following refinement, yielded 12 alerts indicative of steatosis. These alerts, 10 of which had precision ≥80%, captured chemicals with shared features and common toxicological profiles. Each was supported with a clear mechanistic rationale, by means of an extensive literature search, and, where possible, was linked to documented adverse outcome pathways (AOPs) describing steatosis. Ultimately, these were evaluated according to the established principles of uncertainty analysis and were shown to adhere to the findable, accessible, interoperable, and reusable (FAIR) Lite principles for data storage and sharing. The final collection of alerts has potential application in hazard identification, read-across, and screening/prioritization to support tiered assessment strategies targeting hepatic steatosis. To enable usage, a web-based implementation of the steatosis profiler that requires only chemical structures as input is available at https://steatosis-profiler.streamlit.app/.