Uncovering the molecular landscape of young-onset diffuse gastric cancer: A relieff-based feature selection analysis on RNA-Seq data.
Diffuse Gastric Cancer (DGC) is an aggressive subtype with a poor prognosis and a lack of specific biomarkers, representing a critical unmet need in oncology. This study aimed to elucidate the key molecular drivers of DGC by integrating RNA-seq data with a multi-faceted bioinformatics approach.
We analyzed RNA-seq data from young-onset DGC and normal tissues (GSE113255, GSE122401). Machine learning (ML) feature selection (ReliefF algorithm) was used to prioritize genes, followed by protein-protein interaction network analysis to identify hub genes. Their roles were further investigated through Gene Ontology, KEGG pathway analysis, tumor microenvironment immune infiltration, miRNA-regulatory network analysis, transcription factor prediction, and computational drug repurposing analyses.
Our ML‑driven approach identified seven hub genes central to DGC pathogenesis including CCL5, CXCR4, MMP9, FOXP3, IL18, TNFSF11, and TNFSF13B. Among these, three prioritized core hub genes (CXCR4, MMP9, and TNFSF13B) were selected based on statistically significant overexpression and complementary functional roles. Specifically, CXCR4 showed a Fold Change of 3.12 (Log2FC = 1.64, FDR = 0.01), MMP9 exhibited the highest magnitude of upregulation (Fold Change = 16.58, Log2FC = 4.05), and TNFSF13B demonstrated the most statistically significant differential expression (Fold Change = 2.32, Log2FC = 1.21, FDR < 0.000001). High CXCR4 expression was identified as a potential prognostic indicator associated with poorer overall survival in the TCGA‑STAD cohort (HR = 1.5, p = 0.0072). We delineated a core regulatory circuitry where NF‑κB (NFKB1/RELA) masterfully regulates the hub gene network. Drug repurposing analysis nominated several FDA‑approved agents, including the VEGF‑A inhibitor Bevacizumab, which indirectly suppresses CXCR4 and MMP9.
This pilot study establishes a robust integrative framework that synergizes ML with network biology. It nominates CXCR4 and TNFSF13B as candidate therapeutic targets and highlights the potential of ML‑based feature selection for discovering biologically relevant, context‑dependent prognostic indicators in DGC. However, we emphasize that these findings are exploratory and hypothesis‑generating, requiring independent validation through experimental studies and larger cohorts before any clinical translation.
We analyzed RNA-seq data from young-onset DGC and normal tissues (GSE113255, GSE122401). Machine learning (ML) feature selection (ReliefF algorithm) was used to prioritize genes, followed by protein-protein interaction network analysis to identify hub genes. Their roles were further investigated through Gene Ontology, KEGG pathway analysis, tumor microenvironment immune infiltration, miRNA-regulatory network analysis, transcription factor prediction, and computational drug repurposing analyses.
Our ML‑driven approach identified seven hub genes central to DGC pathogenesis including CCL5, CXCR4, MMP9, FOXP3, IL18, TNFSF11, and TNFSF13B. Among these, three prioritized core hub genes (CXCR4, MMP9, and TNFSF13B) were selected based on statistically significant overexpression and complementary functional roles. Specifically, CXCR4 showed a Fold Change of 3.12 (Log2FC = 1.64, FDR = 0.01), MMP9 exhibited the highest magnitude of upregulation (Fold Change = 16.58, Log2FC = 4.05), and TNFSF13B demonstrated the most statistically significant differential expression (Fold Change = 2.32, Log2FC = 1.21, FDR < 0.000001). High CXCR4 expression was identified as a potential prognostic indicator associated with poorer overall survival in the TCGA‑STAD cohort (HR = 1.5, p = 0.0072). We delineated a core regulatory circuitry where NF‑κB (NFKB1/RELA) masterfully regulates the hub gene network. Drug repurposing analysis nominated several FDA‑approved agents, including the VEGF‑A inhibitor Bevacizumab, which indirectly suppresses CXCR4 and MMP9.
This pilot study establishes a robust integrative framework that synergizes ML with network biology. It nominates CXCR4 and TNFSF13B as candidate therapeutic targets and highlights the potential of ML‑based feature selection for discovering biologically relevant, context‑dependent prognostic indicators in DGC. However, we emphasize that these findings are exploratory and hypothesis‑generating, requiring independent validation through experimental studies and larger cohorts before any clinical translation.