Integrative machine learning and multi-omics identify a centromere gene signature and validate B3GALT4 as a tumor suppressor in osteosarcoma.

Centromere-associated genes are linked to genomic instability and tumor progression, yet their significance and immunological roles in osteosarcoma (OS) remain undefined. Methods: We developed a centromere-associated prognostic gene model (CPGM) by applying machine learning across independent cohorts, including TARGET-OS and GEO databases. The Meta-OS and GEO datasets were integrated to form a comprehensive Meta-OS cohort for further analysis. Model performance was evaluated using Harrell's concordance index (C-index). Then, the optimal model was selected and validated using Kaplan-Meier plotter, time-dependent ROC (tROC), Cox regression, and nomogram construction. Functional enrichment and immune infiltration analyses were performed to characterize the tumor immune microenvironment. Single-cell RNA sequencing (scRNA-seq) was conducted to assess cellular heterogeneity and intercellular communication. Finally, the function of Beta-1,3-galactosyltransferase 4 (B3GALT4) was validated using in vitro assays to evaluate its effects on OS cell proliferation and migration. Results: The CPGM demonstrated consistent prognostic performance across independent datasets, achieving a maximum C-index at 0.733. The tROC analysis showed strong predictive accuracy, with area under of curve values of 0.857, 0.806, and 0.783 at 1, 3, and 5 years, respectively. Univariate and multivariate Cox analyses confirmed CPGM as an independent prognostic factor (p = 0.002). Meanwhile, high CPGM scores were associated with poor survival, higher tumor purity, reduced immune infiltration, and suppressed antitumor immune responses. Enrichment analyses indicated significant involvement of immune-related pathways. scRNA-seq analysis revealed that high-CPGM cells were enriched in early developmental trajectories and exhibited enhanced intercellular signaling. Among the model genes, B3GALT4 was consistently downregulated in OS tissues and cell lines, and low expression correlated with poor prognosis. Overexpression of B3GALT4 significantly inhibited the proliferation, migration and invasion of OS cells. Conclusion: This study establishes a robust centromere-associated prognostic model linked to immune dysregulation in OS and identifies B3GALT4 as a potential tumor suppressor and therapeutic target.