Machine Learning-Based Prognostic Model for Gastric Cancer Using Integrated Multi-Omics Data.

Gastric cancer (GC) prognosis remains suboptimally defined by conventional clinicopathological parameters, necessitating integrative multi-omics approaches to unravel molecular heterogeneity. This study established a robust multi-omics prognostic framework through synergistic analysis of transcriptomic, epigenomic, and clinical data from 108 GC patients. Genome-wide expression profiling and methylation array analysis identified 1,243 survival-associated transcripts and 8,742 prognostic CpG sites, with cross-omics integration via similarity network fusion revealing three molecular subtypes exhibiting distinct clinical trajectories. The aggressive Subtype 3 demonstrated a 2.87-fold increased mortality risk compared to the favorable Subtype 1, independent of age and tumor stage. A LASSO-derived prognostic signature integrating eight gene expression markers, nine methylation loci, and three clinical parameters achieved superior discrimination (C-index: 0.786 [95% CI: 0.748-0.824], compared to 0.687-0.752 in unimodal models) and 19-28% improvement in time-dependent AUC metrics. The multi-optimized nomogram incorporating molecular risk scores with conventional predictors demonstrated strong calibration (slope 0.967) and clinical utility across validation cohorts (C-index 0.742), significantly outperforming existing stratification systems. Functional characterization revealed subtype-specific enrichment in cell cycle dysregulation and immune evasion pathways, obtaining CDK/PI3K inhibitors as potential therapeutic targets. These findings establish multi-omics integration as a novel strategy for prognostic refinement and precision therapy guidance in GC.