Integrating single-cell RNA-seq and machine learning to dissect polyamine metabolism in metabolic dysfunction-associated steatotic liver disease

Background Dysregulated immunometabolism is central to the pathogenesis of metabolic dysfunction-associated steatotic liver disease (MASLD). Although polyamines contribute to cellular stress responses and immune-cell function, their cell-type-specific transcriptional associations within the hepatic immune microenvironment remain incompletely understood. Methods We assessed polyamine metabolism in MASLD at the single-cell level using AUCell, UCell, singscore, and AddModuleScore. To find metabolism-related genes, we performed differential analyses. We then combined six machine learning methods—including LASSO, Random Forest, XGBoost, GBM, SVM, and Boruta—to identify and sort robust disease genes. We further evaluated these findings using bulk transcriptomic datasets and a CDAA-induced MASH mouse model. Results We observed notable differences in polyamine metabolic activity among various liver cell types, with relatively higher levels detected in macrophages, cholangiocytes, and stromal cells. PSMB7 and PSMD7 emerged as proteasome-associated candidate genes that were enriched in macrophages and upregulated in MASLD. Higher expression of these genes was associated with immune-related transcriptional programs, including antigen-processing/presentation signatures and predicted intercellular communication pathways involving MIF- and TNFSF13B-related signaling. Their upregulation was further supported by bulk RNA analyses and the CDAA-induced MASH model. Conclusion Our single-cell analysis showed clear heterogeneity in polyamine-metabolism-related states in MASLD, with macrophages emerging as a major associated cell population. PSMB7 and PSMD7 emerged as proteasome-associated candidate markers enriched in macrophage populations with elevated polyamine-metabolism scores, providing a framework for future mechanistic investigation.