Proteomic clocks combined with deep learning phenotypes track eye aging and diseases

Proteomics represents a powerful but underutilized approach for characterizing eye aging. Here, leveraging data from three large-scale, cross-national cohorts of over 55,000 transethnic participants, we demonstrate the ability of high-throughput proteomics combined with deep learning (DL) phenotyping to track eye aging and disease in both discovery and external validation settings. Proteomic aging driven by machine learning modeling closely aligns with signals of eye aging and DL aging phenotypes. We identifiy and validate premature proteomic aging in individuals with major age-related eye diseases (AREDs), including cataract, diabetic retinopathy, age-related macular degeneration, and glaucoma, and propose evidence supporting proteomic aging acceleration as a robust biomarker for predicting these conditions beyond chronological age, with adaptability across sexes and ethnicities. We also develop a streamlined, cost-effective proteomic aging clock that preserves predictive performance while reducing assay complexity. By integrating advanced tomographic and angiographic imaging, we derive structural and functional biomarkers through DL-driven pipelines and link accelerated proteomic aging to both neuroretinal degeneration and microvascular rarefaction in the Guangzhou Diabetic Eye Study (GDES) and the High-definition Oculo-Phenomic Evaluation (HOPE) study, highlighting coupled neural-vascular decline in eye aging. Our findings position proteomic aging combined with AI as a scalable tool for tracking eye health and disease, and provides new insights into shared aging pathways underlying multiple ocular pathologies.