Machine learning based immune evasion signature for predicting the prognosis and immunotherapy benefit in stomach adenocarcinoma.

Background: Stomach adenocarcinoma (STAD) remains a major contributor to cancer-related mortality worldwide. Despite advances in immunotherapy, only a subset of STAD patients benefits from immune checkpoint inhibitors, largely due to tumor-intrinsic immune evasion mechanisms. Therefore, robust predictive biomarkers are urgently needed to guide prognosis assessment and therapeutic decision-making. Methods: An integrative machine learning framework incorporating 10 algorithms was applied to construct an immune evasion signature (IES) using 101 model combinations. The optimal model was selected based on concordance index (C-index) across validation datasets. The prognostic and immunological relevance of the IES was assessed via survival analyses, immune infiltration deconvolution, and multiple immunotherapy response metrics. Key genes were further validated using qPCR, immunohistochemistry, and Results: A four-gene IES developed via the LASSO method demonstrated robust prognostic power across TCGA and multiple external cohorts. High IES score were associated with poor survival, reduced immune cell infiltration (e.g., CD8 Conclusion: We established a novel IES with strong potential to predict prognosis and immunotherapy response in STAD. This IES may serve as a valuable tool for risk stratification and individualized treatment planning in clinical practice.