Cell Type Specific Aging Transcriptional Signatures of Human Retina Through Integrated Machine Learning and Single-Cell Transcriptomics

Abstract

Single-cell RNA sequencing (scRNA-seq) has significantly advanced our understanding of retinal aging, yet the specific molecular characteristics within cell populations remain incompletely defined. We profiled 223,612 single cells from 18 unfrozen human retinas obtained from 13 Chinese donors aged 34-92 years, providing an ethnically specific atlas across the adult lifespan. A sparsity-driven machine-learning (ML) pipeline (L1-regularized logistic regression, recursive feature elimination with cross-validation) identified age-discriminatory genes within each major retinal cell type, complemented by gene set scoring for cellular senescence and metabolic pathways. Using integrated differential expression and ML feature selection, we identified eleven major retinal cell populations and observed aging-associated shifts. ML classifiers achieved high accuracy (80-96%), particularly for microglia (96%), revealing mitochondria-centric aging signatures in rods and bipolar cells, proteostasis and retinoid metabolism in cones, and structural-RNA maintenance signatures in horizontal cells. This study delivers the first ML-derived, cell-type-specific aging gene signatures for the human retina in a Chinese cohort, offering a reference for population-tailored biomarker discovery.