Development and validation of a machine learning-based prognostic model using mitochondrial dysfunction-related genes for colorectal cancer patients.

Background: Colorectal cancer (CRC) represents a primary cause of cancer-related mortality, necessitating novel prognostic biomarkers and therapeutic strategies. Mitochondrial dysfunction, a hallmark of cancer, drives metabolic reprogramming and immune evasion, but its prognostic potential in CRC has not been fully explored. Deep insights and prognostic models related to mitochondrial dysfunction in CRC are currently lacking. This study aimed to develop a machine learning (ML)-based prognostic model using mitochondria-related genes (MRGs) to stratify CRC patients and guide personalized therapy. Methods: RNA sequencing, clinical data from The Cancer Genome Atlas Program containing 473 CRC/41 normal samples, and Gene Expression Omnibus datasets, including GSE39582, GSE38832, and GSE17536, were analyzed. Differential analysis was employed to identify 316 differentially expressed MRGs. Key pathways were screened through functional enrichment analysis. Three ML algorithms and least absolute shrinkage and selection operator regression were utilized to identify prognostic genes. A risk model was developed and validated for overall survival (OS) prediction. Immune infiltration, drug sensitivity, and immunotherapy response were assessed. Results: The 7-gene signature ( Conclusions: This ML-based model using MRGs effectively predicts CRC prognosis, immune microenvironment, and therapeutic response, offering a framework for precision oncology. The 7-gene signature may guide risk stratification and targeted therapy, bridging mitochondrial biology with clinical outcomes.