A Multi-Machine Learning Consensus Agent Identifies Heme Metabolism-Related Prognostic Biomarkers in Ovarian Cancer with Experimental Validation

Ovarian cancer (OC) remains a lethal gynecologic malignancy challenged by chemoresistance and recurrence. Dysregulation of heme metabolism is increasingly implicated in OC progression, yet the underlying mechanisms remain elusive. This study aims to identify robust heme metabolism-related biomarkers and decode their prognostic and therapeutic implications using an advanced artificial intelligence (AI) framework coupled with clinical experimental validation. Differentially expressed genes (DEGs) and heme metabolism-related genes (HMRGs) were integrated across multiple OC cohorts. To overcome the limitations of traditional single-algorithm dimensionality reduction, an autonomous multi-machine learning consensus agent was deployed to rigorously screen and establish the most stable prognostic signature. The biological mechanisms were decoded via gene set enrichment analysis (GSEA) and immune infiltration profiling. Crucially, reverse transcription quantitative polymerase chain reaction (RT-qPCR) was performed on an independent clinical cohort of OC and normal tissues procured from Nanfang Hospital to validate computational predictions. The AI-driven consensus agent distilled 101 candidate genes into a highly robust four-gene signature (EPB42, SLC7A11, HMBS, and GLRX5), which demonstrated excellent predictive efficacy across independent cohorts. RT-qPCR experiments on clinical samples definitively confirmed the significant downregulation of EPB42 and the upregulation of SLC7A11, HMBS, and GLRX5 in OC tissues, perfectly aligning with the in silico findings. Mechanistically, GSEA revealed that high risk is deeply tied to impaired mitochondrial oxidative phosphorylation (OXPHOS), while immune profiling highlighted a coordinated dysregulation of central memory T cells and natural killer (NK) cells. By coupling a novel multi-machine learning consensus agent with real-world clinical experimental validation, this study established a reliable four-gene heme metabolism signature in OC, providing dual computational and experimental evidence linking heme metabolism to mitochondrial dysfunction and tumor immunity, and offering actionable targets for precision oncology.