Machine learning-based construction of immunogenic cell death-related score for improving prognosis and response to immunotherapy in melanoma.

Abstract

Immunogenic cell death (ICD) is a form of regulated cell death (RCD) which could drive the activation of the innate and adaptive immune responses. In this work, we aimed to develop an ICD-related signature to facilitate the assessment of prognosis and immunotherapy response for melanoma patients. A set of machine learning methods, including consensus clustering, non-negative matrix factorization (NMF) method and least absolute shrinkage and selection operator (LASSO) logistic regression model, and bioinformatics analytic tools were integrated to construct an ICD-related risk score (ICDscore). CIBERSORT and ESTIMATE algorithm were used to evaluate the infiltration of immune cells. The 'pRRophetic' package in R and 6 cohorts of melanoma patients receiving immunotherapy were used for therapy sensitivity analyses. The predictive performance between ICDscore with other mRNA signatures were also compared. The ICDscore could predict prognosis and immunotherapy response in multiple cohorts, and displayed superior performance than other forms of cell death-related signatures or 52 published signatures. The melanoma patients with low ICDscore were marked with high infiltration of immune cells, high expression of immune checkpoint inhibitor-related genes, and increased tumor mutation burden. In conclusion, we constructed a stable and robust ICD-related signature for evaluating the prognosis and benefits of immunotherapy, and it could serve as a promising tool to guide decision-making and surveillance for individual melanoma patients.