Integrated machine learning and single-cell analysis reveal the prognostic and therapeutic potential of SUMOylation-related genes in ovarian cancer.

INTRODUCTION: Ovarian cancer (OC) exhibits high mortality and chemoresistance rates, underscoring the urgent need for precise prognostic biomarkers and novel therapeutic targets. SUMOylation, crucial in cellular stress responses, is frequently dysregulated in various cancers. This study aims to characterize SUMOylation and its regulators in OC and identify potential biomarkers and therapeutic targets. METHODS: In this study, using multi-omics data, we characterized the unique features of SUMOylation in OC and revealed the association between SUMOylation-related genes (SRGs) and OC malignancy. We conducted integrated machine learning and single-cell RNA sequencing data analysis to identify key SRGs and explored their functional characteristics. The prognostic potential of these SRGs was confirmed in ID8 mouse models and in samples from 213 OC patients at West China Second Hospital. RESULTS: An integrated machine learning framework identified 22 prognostic-related SRGs from the TCGA-OV cohort. Further single-cell analysis refined these findings, pinpointing five SRGs as biomarkers closely associated with OC cell function, metabolism and the tumor microenvironment. In cancer cells, the expression of four SRGs (PI3, AUP1, CD200 and GNAS) is closely associated with epigenetic regulation and epithelial-mesenchymal signaling. Notably, we found that AUP1 overexpression may contribute to chemoresistance in OC. In the tumor microenvironment, CD8(+) cytotoxic T cell with high CCDC80 (another SRG) expression exhibit inhibited cytotoxicity activity. DISCUSSION: Overall, five SRGs were identified and further evaluated as potential prognostic and therapeutic targets, offering deeper insights into precision oncology for OC.