Machine learning-based identification of an oxidative phosphorylation signature for prognosis, immune infiltration, and drug sensitivity in ovarian cancer.

Background: Ovarian cancer (OC) is a highly heterogeneous disease, and its metabolic characteristics also exhibit heterogeneity. However, the specific metabolic pathways that play a critical role in OC metabolism remain unclear. Additionally, the significance of genes related to the metabolic pathways in the prognosis and therapeutic outcomes has not been clearly defined. Methods: In this study, we utilized the Cancer Genome Atlas Program (TCGA), Genotype-Tissue Expression (GTEx), and multiple Gene Expression Omnibus (GEO) datasets to perform gene set enrichment analysis (GSEA) on 84 metabolic pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG). Through robust rank aggregation (RRA) analysis, we identified the most significantly altered metabolic pathways. By constructing the most robust machine learning model using genes related to the most significantly altered metabolic pathways and combining it with single-cell sequencing analysis results, kinesin family member 1A (KIF1A) was selected as the gene for subsequent biological level studies. Results: We identified oxidative phosphorylation (OXPHOS) as one of the core metabolic pathways in OC. The OXPHOS-related gene signature (OPRGS) was built using the random survival forest (RSF) and supervised principal components (SuperPC) methods, which emerges as a comparatively reliable risk factor for OC. Patients with high-risk scores exhibited higher ESTIMATE stromal-related scores, a significant positive correlation with tumor-associated fibroblasts, higher tumor immune dysfunction and exclusion scores, and lower programmed cell death protein-1 (PD-1) and cytotoxic T lymphocyte-associated antigen-4 (CTLA-4) immunophenoscores in the TCGA cohort, suggesting an immunosuppressive tumor microenvironment (TME) based on bioinformatic predictions. Additionally, higher OPRGS was associated with lower cancer stemness indices, resistance to paclitaxel but sensitivity to carboplatin, revealing complex biological behaviors of the tumor. Further analysis showed that high OPRGS were also correlated with high scores in cancer-related hallmark signaling pathways, such as Notch, angiogenesis, and epithelial-mesenchymal transition signaling pathways. By integrating single-cell RNA sequencing data, we identified KIF1A as a key gene for further investigation. Our findings indicated that KIF1A was upregulated in OC cell lines and might promote cell proliferation, invasion, and migration. Conclusion: This study constructed a new OPRGS for OC. It may serve as a potential indicator for predicting prognosis, immune infiltration, and chemotherapy drug sensitivity in OC patients.