Machine Learning Combined with Fatty Acid Metabolism Genes Identifies a Novel Prognostic Model in Gastric Cancer

Background Gastric cancer (GC) ranks among the most lethal malignancies globally, with marked inter-patient prognostic heterogeneity and an urgent need for reliable molecular biomarkers. Fatty acid metabolism reprogramming is a hallmark of tumor development; however, its specific prognostic utility in GC remains incompletely defined. Methods The GSE66229 discovery dataset (GEO; n = 400: 300 GC tissues, 100 normal gastric mucosae) was analyzed by limma to identify differentially expressed genes (DEGs). Weighted gene co-expression network analysis (WGCNA) isolated GC-relevant co-expression modules; their intersection with a curated 50-gene fatty acid metabolism panel yielded candidate hub genes. Three independent machine-learning algorithms—specifically LASSO regression, random forest, and SVM-RFE—were applied to select core prognostic genes. A risk-score model was constructed and independently validated in the TCGA-STAD cohort (n = 375). Biological mechanisms were explored by gene set variation analysis (GSVA) and single-sample GSEA (ssGSEA). Results A total of 1,265 DEGs were identified. WGCNA revealed two strongly GC-associated modules (blue: r = 0.95, p < 1×10⁻²⁰⁰; turquoise: r = 0.84, p < 1×10⁻²⁰⁰). All three machine-learning algorithms consistently nominated ADIPOQ and FABP4 as hub genes. A strong positive correlation between ADIPOQ and FABP4 (r = 0.835, p ≈ 0) was observed in GEO. In TCGA-STAD, low ADIPOQ expression (p = 0.013) and high FABP4 expression (p = 0.0052) were each independently associated with inferior overall survival (OS). The combined risk score (RiskScore = − 0.5 × Z[ADIPOQ] + 0.8 × Z[FABP4]) was confirmed as an independent prognostic factor by multivariate Cox regression (HR = 0.61, 95% CI: 0.44–0.84, p = 0.0029). Time-dependent AUC values at 1, 3, and 5 years were 0.594, 0.616, and 0.654, respectively—consistently surpassing single clinical variables. Decision curve analysis (DCA) confirmed net clinical benefit. High-risk patients showed enrichment of epithelial-mesenchymal transition (EMT) and angiogenesis pathways, elevated mast cell and B-cell infiltration, and reduced regulatory T-cell (Treg) abundance. Conclusions A robust, biologically grounded prognostic model built on ADIPOQ and FABP4 was developed and validated in GC. The model functions as an independent prognostic indicator and reveals mechanistic links between lipid metabolic reprogramming, invasive phenotype, and immune microenvironment remodeling, providing a novel molecular tool for individualized GC prognosis assessment.