Integrated multi-omics and machine learning highlight PTM-related genes as potential biomarkers in endometrial cancer

Endometrial cancer, the most common gynecological malignancy, represents about 5 % of all cancers in women, with rising incidence rates. Early-stage patients have favorable outcomes, but advanced or recurrent disease leads to poor prognosis and low survival. Current treatments primarily rely on surgery, underscoring the need for new therapeutic targets and biomarkers. Post-translational modifications are crucial in tumor development by regulating protein activity and stability, yet their role in endometrial cancer remains underexplored.

We integrated TCGA and GEO endometrial cancer data. PTM scores were calculated using AddModuleScore, and functional enrichment was conducted with clusterProfiler. Single-cell analysis was done with Seurat and Harmony, followed by UMAP and Louvain clustering. Machine learning techniques, including LASSO regression, SVM, and random forest, were applied to identify key genes. SHAP analysis evaluated their diagnostic contributions, and immune cell infiltration was analyzed using ssGSEA. MR on the FinnGen dataset assessed causal links between eQTLs of six genes and endometrial cancer.

Differential expression analysis identified 2329 downregulated and 2910 upregulated genes in TCGA, and 2274 downregulated and 2545 upregulated genes in GEO. PTM analysis revealed key pathways, including protein deubiquitination and histone modification. Single-cell analysis identified 12 subpopulations. Six key genes were identified with high diagnostic accuracy. SHAP analysis highlighted UBE2C. MR analysis suggested high MYC expression as a risk factor (OR: 1.67, 95 % CI: 1.07-2.60, P = 0.024).

This study provides valuable insights into the role of PTMs in endometrial cancer, identifying potential biomarkers for diagnosis and treatment.