Abstract
BACKGROUND: Gastric adenocarcinoma (GAC) exhibits marked interpatient heterogeneity. Compact, biologically interpretable signatures linked to tumor-immune contexture may aid reproducible risk stratification and hypothesis generation for therapeutic responsiveness. METHODS: TCGA-GAC RNA-seq (n = 407; 375 tumors, 32 normals) was analyzed with limma to identify DEGs (p < 1 × 10⁻³, |log₂FC|>1). ImmPort overlap yielded immune DEGs, which were interaction-expanded via STRING to construct an immune-enriched candidate set for unsupervised co-expression analysis. WGCNA on expressed candidates identified co-expression modules; modules most associated with the tumor phenotype were prioritized for network centrality analysis, survival screening (maxstat), and multivariable Cox modeling to derive an immune-related gene prognostic index (IRGPI). RESULTS: XRCC2, NUSAP1, and ZWILCH were retained as independent prognostic factors and combined into IRGPI (0.526·XRCC2 + 0.581·NUSAP1 - 0.498·ZWILCH). IRGPI significantly stratified overall survival in TCGA with modest discrimination (time-dependent AUC ≈ 0.55) and showed reproducible stratification in GSE26942 (AUC ≈ 0.63). CIBERSORT-based deconvolution identified three TME clusters with distinct survival patterns. IRGPI groups differed in in silico-inferred immunotherapy-related features (TIDE-derived metrics and PD-L1/MDSC/T-cell dysfunction signatures) and in predicted cisplatin/gemcitabine sensitivity by pRRophetic. CONCLUSIONS: This three-gene IRGPI provides a compact, biologically anchored risk stratification signal in GAC and is associated with computationally inferred immune contexture and therapy-related signatures. These associations are hypothesis-generating and require validation in treatment-annotated cohorts, particularly gastric cancer patients receiving immune checkpoint inhibitors; robustness to alternative gene-selection strategies should also be evaluated in future work.