Abstract / Summary
Abstract Defective mitophagy is a key pathological risk of Alzheimer’s disease (AD), yet available pharmacological agents capable of restoring mitophagy remain clinically limited, largely due to the lack of tools for drug screening and target identification. To address this, we introduce an integrated platform with three novel artificial intelligence (AI)-enabled approaches—the phenotype-based screening model ‘CellAutoFM’, the drug-likeness predictor ‘ADMET Ranker’, and the target deconvolution method ‘PROCAT’. Using this platform together with wet lab validation, Crassifolin A (JGX-1) is identified as a potent natural product-derived mitophagy inducer; structurally, JGX-1 directly binds to ATP-binding cassette sub-family A member 1 (ABCA1) and sub-family G member 1 (ABCG1), triggering enhanced cholesterol efflux which activates PINK1/Parkin-dependent mitophagy. In both in vitro and in vivo AD laboratory models, JGX-1 reduces amyloid-β and Tau pathologies in an autophagy/mitophagy-dependent manner as these benefits are dramatically blunted with key mitophagy genes knocked down. Collectively, we provide an open-access, reproducible AI-assisted platform for rapid compound screening, drug-likeness evaluation, and potential target prediction to advance drug discovery. Through this paradigm, we unveil modulating the ABCA1/ABCG1-mediated cholesterol-mitophagy pathway as a promising anti-AD strategy, with a novel drug candidate JGX-1 identified.