Investigating combined hypoxia and stemness indices for prognostic transcripts in gastric cancer: Machine learning and network analysis approaches

Abstract

Introduction

Gastric cancer (GC) is among the deadliest malignancies globally, characterized by hypoxia-driven pathways that promote cancer progression, including stemness mechanisms facilitating invasion and metastasis. This study aimed to develop a prognostic decision tree using genes implicated in hypoxia and stemness pathways to predict outcomes in GC patients.

Materials and methods

GC RNA-seq data from The Cancer Genome Atlas (TCGA) were analyzed to compute hypoxia and stemness scores using Gene Set Variation Analysis (GSVA) and the mRNA expression-based stemness index (mRNAsi). Hierarchical clustering identified clusters with distinct survival outcomes, and differentially expressed genes (DEGs) between clusters were identified. Weighted Gene Co-expression Network Analysis (WGCNA) identified modules and hub genes associated with clinical traits. Overlapping DEGs and hub genes underwent functional enrichment, protein-protein interaction (PPI) network analysis, and survival analysis. A prognostic decision tree was constructed using survival-associated shared genes.

Results

Hierarchical clustering identified six clusters among 375 TCGA GC patients, with significant survival differences between cluster 1 (low hypoxia, high stemness) and cluster 4 (high hypoxia, high stemness). Validation in the GSE62254 dataset corroborated these findings. WGCNA revealed modules linked to clinical traits and survival, with functional enrichment highlighting pathways like cell adhesion and calcium signaling. The decision tree, based on genes such as AKAP6, GLRB, and RUNX1T1, achieved an AUC of 0.81 (training) and 0.67 (test), demonstrating the utility of combined scores in patient stratification.

Conclusion

This study introduces a novel hypoxia-stemness-based prognostic decision tree for GC. The identified genes show promise as prognostic biomarkers, warranting further clinical validation.