Machine learning-based integration of tumor deposit molecular signatures improves prognostic stratification in colon adenocarcinoma

BACKGROUND: Colon adenocarcinoma (COAD) remains a leading cause of cancer-related mortality worldwide. Although tumor deposits (TDs) are established prognostic indicators, their molecular characteristics and potential for improving risk stratification remain unexplored. METHODS: We performed an integrative analysis of transcriptomic and clinical data from TCGA and GEO databases to identify TD-associated molecular signatures. A hybrid ML framework combining random survival forest and stepwise Cox regression was developed to construct a risk stratification model. Model performance was validated through survival analysis, time-dependent ROC curves, and multivariate analyses. Gene set enrichment analysis explored underlying mechanisms and therapeutic implications. RESULTS: The integrated molecular signature-based model demonstrated superior prognostic accuracy, effectively stratifying patients into risk groups with distinct survival outcomes (P < 0.001) and clinicopathological features. High-risk patients exhibited enhanced immune evasion mechanisms and differential drug sensitivity patterns. Pathway analysis revealed significant alterations in ECM receptor interaction, PPAR signaling, and neuroactive ligand-receptor interaction pathways. CONCLUSIONS: Our machine learning-based integration of TD molecular signatures establishes a robust risk stratification model for COAD patients, offering improved prognostic accuracy and valuable insights for personalized treatment strategies. Our findings highlight the potential of interpretable machine learning in molecular oncology risk modeling.