Machine learning developed an intratumor heterogeneity signature for predicting prognosis and immunotherapy benefits in skin cutaneous melanoma.

Abstract

Intratumor heterogeneity (ITH) is defined as differences in molecular and phenotypic profiles between different tumor cells and immune cells within a tumor. ITH was involved in the cancer progression, aggressiveness, therapy resistance and cancer recurrence. Integrative machine learning procedure including 10 methods was conducted to develop an ITH-related signature (IRS) in The Cancer Genome Atlas (TCGA), GSE54467, GSE59455 and GSE65904 cohort. Several scores, including tumor immune dysfunction and exclusion (TIDE) score, tumor mutation burden (TMB) score and immunophenoscore (IPS), were used to evaluate the role of IRS in predicting immunotherapy benefits. Two immunotherapy datasets (GSE91061 and GSE78220) were utilized to the role of IRS in predicting immunotherapy benefits of skin cutaneous melanoma (SKCM) patients. The optimal prognostic IRS constructed by Lasso method acted as an independent risk factor and had a stable and powerful performance in predicting the overall survival rate in SKCM, with the area under the curve of 2-, 3- and 4-year receiver operating characteristic curve being 0.722, 0.722 and 0.737 in TCGA cohort. We also constructed a nomogram and the actual 1-, 3- and 5-year survival times were highly consistent with the predicted survival times. SKCM patients with low IRS scores had a lower TIDE score, lower immune escape score and higher TMB score, higher PD1&CTLA4 IPS. Moreover, SKCM patients with low IRS scores had a lower gene sets score involved in DNA repair, angiogenesis, glycolysis, hypoxia, IL2-STAT5 signaling, MTORC1 signaling, NOTCH signaling and P53 pathway. The current study constructed a novel IRS in SKCM using 10 machine learning methods. This IRS acted as an indicator for predicting the prognosis and immunotherapy benefits of SKCM patients.