A robust ammonia metabolism gene signature identified by machine learning predicts prognosis and immunotherapy response in clear cell renal cell carcinoma

Background Despite the widespread use of immune checkpoint inhibitors (ICIs) in advanced clear cell renal cell carcinoma (ccRCC), therapeutic resistance persists. The prognostic and immunomodulatory role of ammonia metabolism remains unclear. Methods We leveraged public RNA-seq data and machine learning to identify ammonia metabolism pathways through enrichment analysis of programmed cell death-related genes. Employing multi-omics data from ccRCC patients, we developed an ammonia metabolism risk score (AMRS) via machine learning, which was validated externally and in immunotherapy cohorts. Additionally, scRNA-seq, WGCNA, TMB analysis, and in vitro assays were performed to characterize the model’s functional basis. Results From 147 prognostic ammonia metabolism-related genes in TCGA, a 4-gene random forest model was constructed using LASSO and multivariate Cox regression. This model demonstrated robust predictive accuracy in external validation (3/5/7-year AUCs: 0.710/0.721/0.771). High-risk patients showed significantly elevated mortality in external cohorts (HR = 4.23, 95% CI 1.57–11.42, p = 0.002) and multiple ICI cohorts (HR = 1.30–1.69, p < 0.05). Functional validation via CSAD-targeted siRNA knockdown suppressed migration and invasion by >44% (p < 0.05) across four ccRCC cell lines. Conclusions Our integrated approach overcomes modeling constraints from limited samples and high-dimensional data and establishes a novel ammonia metabolism-related prognostic signature for ccRCC. CSAD emerges as a promising biomarker warranting further investigation.