Abstract
BACKGROUND: Programmed cell death (PCD) genes play crucial roles in cancer progression and response to therapies, yet their impact on gastric cancer remains inadequately elucidated. This study aimed to create a prognostic cell death signature (PCDs) for gastric cancer, providing insights into potential therapeutic targets and survival predictors. METHODS: We utilized TCGA-STAD and five GEO datasets, representing thousands of gastric cancer samples, for a comprehensive analysis of PCD genes. Differential gene expression, functional enrichment, survival, and machine learning analyses were conducted to construct a PCD-based prognostic model. RESULTS: A total of 249 differentially expressed PCD genes were identified between cancerous and noncancerous gastric tissues. Subsequently, a PCD signature based on seven genes was developed and cross-validated across multiple cohorts. The high-PCD subtype correlated with poorer survival outcomes, lower tumor mutational burden, higher infiltration of M2 macrophages, lower levels of immune checkpoint expression, and decreased response to immunotherapy. A nomogram incorporating the PCDs provided accurate survival rate predictions. Additionally, nine machine learning algorithms were implemented for recurrence prediction, with the random forest model displaying high effectiveness. In this model, thrombospondin 1 (THBS1) showed the highest weight, and its knockdown significantly reduced gastric cancer cell proliferation and invasion. CONCLUSION: This study underscores the significance of PCD genes, particularly THBS1, in gastric cancer progression and highlights their value as potential therapeutic targets. The predictive models developed here can aid in assessing patient prognosis and tailoring personalized treatment strategies.