Programmed cell death signatures-driven microglial transformation in Alzheimer's disease: single-cell transcriptomics and functional validation.

Background: This study aims to develop and validate a programmed cell death signature (PCDS) for predicting and classifying Alzheimer's disease (AD) using an integrated machine learning framework. We further explore the role of S100A4 in AD pathogenesis, particularly in microglia. Methods: A total of one single-cell RNA sequencing (scRNA-seq) and four bulk RNA-seq datasets from multiple GEO datasets were analyzed. Weighted Gene Co-expression Network Analysis (WGCNA) was utilized to identify PCD-related genes. An integrated machine learning framework, combining 12 algorithms was used to construct a PCDS model. The performance of PCDS was validated using multiple independent cohorts. Results: ScRNA-seq analysis revealed higher PCD levels in microglia from AD patients. Seventy-seven PCD-related genes were identified, with 70 genes used to construct the PCDS model. The optimal model, combining Stepglm and Random Forest, achieved an average AUC of 0.832 across five cohorts. High PCDS correlated with upregulated pathways related to inflammation and immune response, while low PCDS associated with protective pathways. Conclusion: This study developed a robust PCDS model for AD prediction and identified S100A4 as a potential therapeutic target. The findings highlight the importance of PCD pathways in AD pathogenesis and provide new insights for early diagnosis and intervention.