A Mitoxyperilysis‐Related Single‐Cell and Machine‐Learning Framework Defines an Immune‐Cold Melanoma Phenotype and a Robust Prognostic Signature

Mitoxyperilysis is a mitochondria‐dependent membrane lysis process driven by innate immune and metabolic cues, yet its clinical relevance in melanoma remains unclear. We analyzed single‐cell RNA‐seq data (GSE215120) to quantify a mitoxyperilysis‐related score (MRS), resolve cell‐type heterogeneity, and compare predicted cell–cell communication between MRS‐high and MRS‐low tumor states. MRS was robust across alternative scoring approaches and varied markedly across cell types and malignant subpopulations. Compared with MRS‐low tumors, the MRS‐high state exhibited increased predicted intercellular communication (740 vs. 448 interactions) and higher global interaction strength (18,989 vs. 10,652), suggesting a rewired tumor ecosystem. To translate these programs to bulk melanoma, we selected the top 150 genes most correlated with MRS and benchmarked 101 machine‐learning strategies in TCGA‐SKCM to derive prognostic models, followed by external validation in six independent GEO cohorts. A gradient boosting machine (GBM)–based signature showed the most consistent cross‐cohort performance and reliably stratified overall survival. High riskScore was associated with reduced immune and stromal signals, higher tumor purity, and an immune‐cold tumor microenvironment as estimated by multialgorithm deconvolution and ESTIMATE. As a representative model gene, GPR143 was upregulated in melanoma, was associated with worse survival, and its functional knockdown suppressed colony formation in melanoma cells. Collectively, this work establishes a novel integrative framework that—for the first time—connects single‐cell–resolved mitoxyperilysis‐associated transcriptional programs with large‐scale multicohort machine‐learning validation, thereby enabling both mechanistic interpretation of immunometabolic heterogeneity and clinically applicable risk stratification in melanoma.