Identification of prognostic biomarkers associated with T and melanoma cell subpopulations in melanoma through integrating machine learning and multiomics.

backgroundMelanoma is a highly lethal cancer with a poor prognosis. T-cells and melanoma cells play crucial role in shaping the tumor microenvironment, yet their role and impact on prognosis in melanoma is still unclear.

methodsWe analyzed single-cell RNA sequencing (scRNA-seq) data for melanoma (GSE200218 and GSE215121) from the Gene Expression Omnibus (GEO) and gene expression data from GSE65904 and TCGA-Skin Cutaneous Melanoma (SKCM). We correlated cells with survival outcomes to identify cell subpopulations linked to melanoma prognosis using Scissor. Based on 108 prognostic genes, melanoma patients were stratified into two subgroups. A novel prognostic risk score (PRS) model was constructed using differentially expressed genes (DEGs) from these subgroups.

resultsOur analysis revealed specific T-cell and melanoma subpopulations influencing melanoma prognosis, validated in an independent cohort. Notably, our study was the first to identify MITF + T-cell and M2-cell sub-populations associated with melanoma prognosis. Using 108 prognostic gene markers, we stratified TCGA-SKCM patients into two groups with distinct clinical outcomes, immune cell scores, and carcinogenic profiles. Additionally, we employed 72 machine-learning algorithm combinations to develop a consensus prognosis model based on 174 DEGs from the two prognosis-related subgroups. Ultimately, we created a novel PRS model using 11 genes, which demonstrated accurate prognostic predictive ability in the GSE65904 validation cohort.

conclusionsThis study identified MITF + T-cells and M2-cells as key factors in melanoma prognosis and developed a novel PRS model for accurate prediction. These findings could help guide clinical decision-making for melanoma patients.