SUMOylation-related genes define prognostic subtypes in stomach adenocarcinoma: integrating single-cell analysis and machine learning analyses.

BACKGROUND: Stomach adenocarcinoma (STAD) exhibits high molecular heterogeneity and poor prognosis, necessitating robust biomarkers for risk stratification. While SUMOylation, a post-translational modification, regulates tumor progression, its prognostic and immunological roles in STAD remain underexplored. METHODS: Prognostic SUMOylation-related genes (SRGs) were screened via univariate Cox regression, and patients were stratified into molecular subtypes using unsupervised consensus clustering. A SUMOylation Risk Score (SRS) model was developed using 69 machine learning models across 10 algorithms, with performance evaluated by C-index and AUC. Immune infiltration, pathway enrichment identified key SRGs, and in vitro functional assays were validated. RESULTS: Two molecular subtypes (A/B) with distinct SUMOylation patterns, survival outcomes (log-rank p < 0.001), and immune microenvironments were identified. The random survival forest (RSF)-based SRS model (AUC: 0.97) stratified patients into high-/low-risk groups, where high-risk patients exhibited advanced tumor stages, immune suppression, and elevated TIDE scores (p < 0.001). Functional enrichment linked low-risk groups to genome stability pathways (DNA repair, cell cycle control). In vitro validation confirmed that L3MBTL2 and VHL knockdown promoted proliferation, migration, and invasion in AGS cells (p < 0.05). CONCLUSION: This study establishes SRGs as independent prognostic indicators and defines SUMOylation-driven subtypes with distinct immune and molecular features. The SRS model and functional validation of L3MBTL2/VHL provide actionable insights for personalized STAD management and immunotherapy targeting. (214 words).