Abstract
BACKGROUND: Glioma is the most frequent malignant primary brain tumor, and mitochondria may influence the progression of glioma. The aim of this study was to analyze the role of nuclear mitochondria related genes (MTRGs) in glioma, identify subtypes and construct a prognostic model based on nuclear MTRGs and machine learning algorithms. METHODS: Samples containing both gene expression profiles and clinical information were retrieved from the TCGA database, CGGA database, and GEO database. We selected 16 nuclear MTRGs and identified two clusters of glioma. Prognostic features, microenvironment, mutation landscape, and drug sensitivity were compared between the clusters. A prognostic model based on multiple machine learning algorithms was then constructed and validated by multiple datasets. RESULTS: We observed significant discrepancies between the two clusters. Cluster One had higher nuclear MTRG expression, a lower survival rate, and higher immune infiltration than Cluster Two. For the two clusters, we found distinct predictive drug sensitivities and responses to immune therapy, and the infiltration of immune cells was significantly different. Among the 22 combinations of machine learning algorithms we tested, LASSO was the most effective in constructing the prognostic model. The model's accuracy was further verified in three independent glioma datasets. We identified MGME1 as a vital gene associated with infiltrating immune cells in multiple types of tumors. CONCLUSION: In short, our research identified two clusters of glioma and developed a dependable prognostic model based on machine learning methods. MGME1 was identified as a potential biomarker for multiple tumors. Our results will contribute to precise medicine and glioma management.