Deep contrastive learning framework identifies cell‐type‐specific drug targets in Alzheimer's disease

Abstract INTRODUCTION Identifying disease‐modifying drug targets is crucial for developing effective Alzheimer's disease (AD) treatments. METHODS We present a deep contrastive learning framework for cell type‐specific AD‐associated genes identification (alzCL). alzCL creates cell‐type‐specific representations of genes by integrating human brain single‐nucleus RNA‐sequencing data with the human protein–protein interactome, thereby capturing both genetic signatures and functional features. RESULTS By integrating human brain snRNA‐seq data, alzCL outperforms the state‐of‐the‐art models by 18% to 24% in area under the receiver operating characteristic curve. Via alzCL, we computationally identified 16, 164, and 221 AD‐associated genes across astrocytes, microglia, and inhibitory neurons, respectively. Top prioritized genes (e.g., MAP3K5, P2RX4 , and PRKD1 ) are significantly enriched in multiple AD‐associated inflammatory and other pathobiological pathways. By integrating drug–target interaction data with alzCL‐predicted AD‐associated genes, we identified potential repurposable drugs for AD, including selonsertib and paroxetine. DISCUSSION AlzCL offers a deep contrastive learning framework for discovery of disease‐associated genes and drug targets in AD.