Identification and validation of a metabolic-related gene risk model predicting the prognosis of Lung, Colon, and Breast cancers

Abstract Metabolic reprogramming, vital for cancer cells to adapt to the altered microenvironment, remains a topic requiring further investigation for different tumor types. Our study aims to elucidate shared metabolic reprogramming across breast (BRC), colorectal (CRC), and lung (LUC) cancers. Leveraging gene expression data from the Gene Expression Omnibus and various bioinformatics tools like MSigDB, WebGestalt, String, and Cytoscape, we identified key/hub metabolism-related genes (MRGs) and their interactions. The functional characteristics including survival parameters and expression of the key MRGs were analyzed and validated through Gene Expression Profiling Interactive Analysis 2 and qRT-PCR. In addition, we employed machine learning algorithms such as k-nearest neighbours (KNN), support vector regressor (SVR), and extreme gradient boosting (XGBoost) to assess MRGs' effectiveness in predicting overall patient survival. Among 11,384 DEGs analyzed, 540 overlapped across BRC, CRC, and LUC, with 46 MRGs and 20 key/hub MRGs involved in all studied cancer types. Of these, 11 key MRGs were prognostically significant. qRT-PCR validation of key MRGs in specific cancer cell lines confirmed their expression profiles, with some showing cell-type-specific patterns. SVR exhibited remarkable accuracy in predicting overall survival, emphasizing its clinical utility. Our integrated approach combining bioinformatics analyses and experimental validations underscores the potential of MRGs as biomarkers for metabolic therapies, with machine learning models enhancing predictive capabilities for patient outcomes.