Machine Learning-based Classifier to Decipher Immune Landscape of Uveal Melanoma and Predict Patient Outcomes

Abstract

Uveal melanoma (UVM) is influenced by immune infiltration features, making the analysis of UVM genomic and immune signatures crucial for predicting patient prognosis and identifying potential targeted therapies.To address this issue, we leveraged multi-omics data from The Cancer Genome Atlas and GEO datasets, especially immune infiltration data, to classify UVM into distinct immune-related subgroups using an unsupervised clustering algorithm. The resulting subgroups were denoted as uveal melanoma carcinoma subtype 1 (UMCS1) and subtype 2 (UMCS2). We further examined differences in the immune microenvironment, immunotherapy response, and tumor metabolic pathways between these subgroups, aiming to identify targets related to immune infiltration. Additionally, we devised a risk scoring system based on subtype-specific markers to forecast the prognosis of UVM patients. Performance evaluation of the risk scoring system was conducted using receiver operating characteristic (ROC) curves, decision curve analysis (DCA), and calibration curves.Our analysis successfully identified two distinct subtypes of UVM patients, characterized by genomic mutations and disparities in the immune environment. These subtypes exhibited diverse clinical features and biological processes. The aggressive subtype, UMCS2, presented a higher TNM stage and poorer patient survival. UMCS2 was distinguished by elevated metabolism and increased immune infiltration. However, UMCS2 also demonstrated a higher tumor mutational burden and immune dysfunction, resulting in diminished responsiveness to immunotherapy. Notably, the two subgroups exhibited differential sensitivity to targeted drugs due to substantial variances in metabolic and immune environments, with UMCS2 displaying lower sensitivity. Finally, we developed a risk scoring system utilizing subtype-specific biomarkers and assessed its diagnostic performance for UVM patients, achieving satisfactory results through ROC curves, decision curve analysis, and calibration curves. Our findings suggest that the remodeled immunometabolic pathways and the immune microenvironment contribute to the relatively low sensitivity of UVM to immunotherapy. Targeting these mutated pathways and immune infiltrating molecules may potentially address the current treatment dilemma in UVM. Moreover, the newly developed risk assessment system not only aids in predicting patient prognosis but also facilitates the identification of suitable populations for combination therapy.