Abstract: Objective To identify core genes influencing immunotherapy response in gastric cancer and to establish a corresponding predictive model. Methods Gene expression data from gastric cancer patients treated with immunotherapy were obtained from the GEO and the TIDE databases.Differentially expressed genes were screened,and weighted gene co-expression network analysis(WGCNA)was performed,identifying 185 genes associated with immunotherapy response for gastric cancer.Core genes were further selected via three machine-learning algorithms:Least Absolute Shrinkage and Selection Operator(LASSO),Random Forest(RF),and Support Vector Machine-Recursive Feature Elimination(SVM-RFE).A predictive model was constructed based on these core genes and validated externally.Model performance was assessed using ROC curves.Additional validation was carried out on an independent dataset GSE62254.Single-cell RNA sequencing was employed to examine the expression of core genes across cell subpopulations. Results A predictive model comprising five core genes was developed,which stratified a cohort of 300 gastric cancer patients into high- and low-risk subgroups.The high-risk group exhibited higher tumor-microenvironment infiltration,TIDE scores,T-cell dysfunction,cancer-associated fibroblast activity,and expression of specific immune-checkpoint molecules,together with lower microsatellite instability scores,collectively indicating enhanced immune escape potential and poorer response to immune checkpoint inhibitors.Single-cell analysis confirmed heterogeneous expression patterns of the five core genes among different cell types. Conclusion A robust predictive model based on differentially expressed core genes related to gastric cancer immunotherapy response was constructed.This model may offer potential value for future research and could aid in clinical decision-making.

Key words: immunotherapy, gastric cancer, machine learning