Comprehensive Summary
The study introduces a metabolic-related gene (MRG)-based prognostic model to predict survival outcomes and immunotherapy responses in cutaneous melanoma (SKCM). Using transcriptomic data from TCGA and GEO datasets (GSE4467, GSE65904), the authors applied 10 machine learning algorithms in a cross-validation framework to identify key metabolic genes associated with prognosis. A six-gene risk signature-SLC27A2, HSD11B1, QPRT, NME1, MICAL2, and GALNT2-was identified and validated. High-risk patients exhibited poorer overall survival (HR = 3.06, p < 0.001), higher metabolic activity, and reduced immune cell infiltration than low-risk patients. The Random Survival Forest (RSF) + StepCox model achieved the highest predictive accuracy (C-index = 0.629). Functional enrichment revealed that the high-risk group was enriched for pathways in angiogenesis, epithelial-mesenchymal transition (EMT), and cell cycle dysregulation, indicating aggressive tumor behavior.
Outcomes and Implications
This research highlights the metabolic reprogramming as a key driver of melanoma progression and introduces GALNT2 as a potential therapeutic target. Experimental assays confirmed that GALNT2 overexpression promotes proliferation and migration in melanoma cell lines (A375, A2058), while its silencing suppresses these malignant traits. The study found that low-risk patients had higher immune and stromal scores and responded better to immune checkpoint inhibitors, emphasizing the relevance of metabolism-immunity cross-talk in melanoma biology. This MRG-based risk model could serve as a precision medicine tool for stratifying patients by prognosis and guiding immunotherapy decisions. By integrating multi-omics data and AI-based modeling, this work advances personalized oncology, underscoring how machine learning can decode metabolic signatures to improve melanoma treatment outcomes.