Machine learning-based identification of an immunotherapy-related signature to enhance outcomes and immunotherapy responses in melanoma.

Abstract

Immunotherapy has revolutionized skin cutaneous melanoma treatment, but response variability due to tumor heterogeneity necessitates robust biomarkers for predicting immunotherapy response. We used weighted gene co-expression network analysis (WGCNA), consensus clustering, and 10 machine learning algorithms to develop the immunotherapy-related gene model (ITRGM) signature. Multi-omics analyses included bulk and single-cell RNA sequencing of melanoma patients, mouse bulk RNA sequencing, and pathology sections of melanoma patients. We identified 66 consensus immunotherapy prognostic genes (CITPGs) using WGCNA and differentially expressed genes (DEGs) from two melanoma cohorts. The CITPG-high group showed better prognosis and enriched immune activities. DEGs between CITPG-high and CITPG-low groups in the TCGA-SKCM cohort were analyzed in three additional melanoma cohorts using univariate Cox regression, resulting in 44 consensus genes. Using 101 machine learning algorithm combinations, we constructed the ITRGM signature based on seven model genes. The ITRGM outperformed 37 published signatures in predicting immunotherapy prognosis across the training cohort, three testing cohorts, and a meta-cohort. It effectively stratified patients into high-risk or low-risk groups for immunotherapy response. The low-risk group, with high levels of model genes, correlated with increased immune characteristics such as tumor mutation burden and immune cell infiltration, indicating immune-hot tumors with a better prognosis. The ITRGM's relationship with the tumor immune microenvironment was further validated in our experiments using pathology sections with GBP5, an important model gene, and CD8 IHC analysis. The ITRGM also predicted better immunotherapy response in eight cohorts, including urothelial carcinoma and stomach adenocarcinoma, indicating broad applicability. The ITRGM signature is a stable and robust predictor for stratifying melanoma patients into 'immune-hot' and 'immune-cold' tumors, enhancing prognosis and response to immunotherapy.