HP-MoleQ: An Effective Predictive Model for High-Throughput Screening of Food-Derived Hepatoprotective Compounds.

Abstract

Hepatoprotective natural compounds are pivotal constituents for the formulation of functional foods and the prophylaxis of hepatic disorders. To enhance the efficiency and reliability of hepatoprotective compound discovery, we propose a new predictive framework named HP-MoleQ. It comprises a pretrained Transformer and an uncertainty quantification module. HP-MoleQ outperforms existing models and enables high-confidence screening of the FoodB database. Compounds are ranked according to 95% confidence interval width lower bound. Top 20 candidates, including 12 phenolic glucopyranosides, 5 phenolic furanosides and 3 flavonoid glycosides, were prioritized. Three candidates (1,2,6-trigalloyl-beta-D-glucopyranose, hamamelitannin, and 6-hydroxyluteolin 7-glucoside) were represented and chosen for experimental validation. In cellular models of alcohol-induced hepatotoxicity and oleic acid and palmitic acid-induced steatosis, the three candidates attenuated intracellular triglyceride, aspartate aminotransferase, and alanine transaminase, inhibited lipid droplet formation, and enhanced albumin production. These findings not only confirm the hepatoprotective efficacy of these compounds but also highlight the utility of the HP-MoleQ platform in identifying potential hepatoprotective agents.