Ganghua Zhang, Guanjun Chen, Jianing Fang, Yiqi Tan, Wenzhi Deng, Zhijing Yin, Ziwei Yin, Jingxin Yang, Le Zeng, Jingyu Ou, Xiangyang Zeng, Biyao Jiang, Songshu Xiao, Yuxing Zhu, Ke Cao
High tumor heterogeneity and platinum resistance are key drivers of poor prognosis in ovarian cancer (OV), yet platinum resistance heterogeneity within the OV tumor microenvironment remains incompletely characterized. Using five algorithms (AUCell, UCell, singscore, ssGSEA, AddModuleScore), we calculated a single-cell platinum resistance score (PRS) and stratified malignant cells into three PRS-based subtypes, followed by intercellular communication and pseudotime trajectory analyses. We identified 12 hub genes of highly resistant malignant (HRM) cells via integrated bioinformatics and machine learning, and constructed the Nebo classification model using the mlr3 framework. The model showed robust predictive performance across multiple single-cell datasets and clinical cohorts. HRM cells exhibited enhanced cisplatin resistance, stemness, intercellular crosstalk, and elevated oncogenic and metabolic pathway activity. SPINT2 was the most upregulated hub gene in platinum-resistant OV cell lines; in vitro and in vivo validation confirmed that SPINT2 knockdown sensitizes OV cells to cisplatin by reducing RAD51 protein stability and inhibiting homologous recombination repair. This study establishes the Nebo model for HRM detection and validates SPINT2 as a promising target to reverse platinum resistance, offering novel insights into OV heterogeneity and therapeutic development.