Machine Learning Models for High-Throughput Screening of Obesogens Based on Pathway Network

Metabolism-disrupting chemicals increase the incidence of obesity through diverse signaling pathways. It is critical to prioritize potential obesogenic factors and systematically profile their modes of action. Here, based on adverse outcome pathway networks (AONs), we established a machine learning screening system that integrated three adverse outcome pathways (AOPs) to describe the mechanisms of chemical-induced obesity for thousands of chemicals. Notably, optimal AON-informed machine learning models achieved an accuracy above 0.95 and identified amide, aromatic/polycyclic, and nitrogen-containing structures as obesity-related. In addition, polycyclic aromatic hydrocarbons were identified as prime candidate obesogens, inducing metabolic syndrome by interfering with all three signaling pathways across the AONs. Representative obesogens predicted in this study, such as antibiotics, insecticide metabolites, and exogenous chemical glucagon, were found to predominantly target membrane, mitochondrial, and nuclear receptor signaling pathways, respectively. Furthermore, top-prioritized chemicals (such as phenanthrene, azithromycin, 1,8,9-trihydroxyanthracene, etc.) according to ranking scores were randomly selected for experimental verification, and all selected candidates exhibited pathway interference in activating specific molecular initiating events (MIEs), demonstrating the predictive model's accuracy. The developed high-throughput screening strategy can efficiently evaluate the potential obesogenic effects of emerging chemicals and provide guidance for the safe design of new chemicals.