Integration of machine learning and experimental validation reveals new lipid-lowering drug candidates.

Abstract

Hyperlipidemia, a major risk factor for cardiovascular diseases, is associated with limitations in clinical lipid-lowering medications. Drug repurposing strategies expedite the research process and mitigate development costs, offering an innovative approach to drug discovery. This study employed systematic literature and guidelines review to compile a training set comprising 176 lipid-lowering drugs and 3254 non-lipid-lowering drugs. Multiple machine learning models were developed to predict the lipid-lowering potential of drugs. A multi-tiered validation strategy was implemented, encompassing large-scale retrospective clinical data analysis, standardized animal studies, molecular docking simulations and dynamics analyses. Through a comprehensive screening analysis utilizing machine learning, 29 FDA-approved drugs with lipid-lowering potential were identified. Clinical data analysis confirmed that four candidate drugs, with Argatroban as the representative, demonstrated lipid-lowering effects. In animal experiments, the candidate drugs significantly improved multiple blood lipid parameters. Molecular docking and dynamics simulations elucidated the binding patterns and stability of candidate drugs in interaction with related targets. We successfully identified multiple non-lipid-lowering drugs with lipid-lowering potential by integrating state-of-the-art machine learning techniques with multi-level validation methods, thereby providing new insights into lipid-lowering drugs, establishing a paradigm for AI-based drug repositioning research, and expanding the repertoire of lipid-lowering medications available to clinicians.