Machine Learning-Based single-cell characterization of lipid metabolic reprogramming in prostate cancer

Prostate cancer (PCa) is one of the most prevalent malignancies in men worldwide, and metabolic reprogramming particularly of lipid metabolism plays a pivotal role in its initiation and progression. Lipid biosynthesis and cholesterol regulation are closely linked to androgen receptor signaling and therapeutic resistance. However, the molecular mechanisms underlying lipid metabolic heterogeneity in prostate cancer remain poorly understood. We integrated multi-omics, single-cell RNA sequencing (scRNA-seq), and spatial transcriptomic data to characterize lipid metabolism-related gene expression in prostate cancer. Public datasets, including TCGA-PRAD and GSE206962, were analyzed using R-based bioinformatics pipelines. Machine learning algorithms, including LASSO-Cox regression and regularized Cox modeling, were applied to identify prognostic lipid metabolism-associated hub genes. Functional enrichment, pseudotime trajectory, and pathway analyses were performed to explore lipid metabolic reprogramming and its association with clinical outcomes. Single-cell transcriptomic analysis revealed marked heterogeneity in lipid metabolic activity across prostate cancer cell populations, with epithelial and endothelial clusters exhibiting the highest metabolic scores. Key hub genes HMGCR, MVK, STARD3, FADS1, and APOE were identified as central regulators of fatty acid and cholesterol metabolism. Spatial transcriptomic mapping showed region-specific enrichment of fatty acid biosynthesis and β-oxidation pathways within tumor tissue. From this module, five genes demonstrated independent prognostic value in the multivariate LASSO-Cox analysis and were subsequently used to construct the optimized prognostic signature. Thus, the 5-gene WGCNA-derived lipid metabolism module panel represents the full pool of biologically relevant candidates, whereas the 5-gene subset constitutes the final risk model. This integrative analysis demonstrates that lipid metabolism reprogramming is a major driver of prostate cancer progression and therapeutic resistance. Machine learning and single-cell approaches revealed key lipid metabolism-related biomarkers with strong prognostic potential. Targeting lipid biosynthesis and cholesterol pathways may provide new opportunities for personalized therapy and improved disease management in prostate cancer.