Abstract
BACKGROUND: Distinctive heterogeneity characterizes diffuse large B-cell lymphoma (DLBCL), one of the most frequent types of non-Hodgkin's lymphoma. Mitochondria have been demonstrated to be closely involved in tumorigenesis and progression, particularly in DLBCL. OBJECTIVE: The purposes of this study were to identify the prognostic mitochondria-related genes (MRGs) in DLBCL, and to develop a risk model based on MRGs and machine learning algorithms. METHODS: Transcriptome profiles and clinical information were obtained from the Gene Expression Omnibus (GEO) database. The risk model was defined using Least Absolute Shrinkage and Selection Operator (Lasso) regression algorithm, and its prognostic value was further examined in independent datasets. Patients were stratified into two clusters based on the risk scores, additionally a nomogram was generated based on the risk score and clinical characteristics. Gene pathway level, microenvironment, expression of targeted therapy-associated genes, response to immunotherapy, drug sensitivity, and somatic mutation status were compared between clusters. RESULTS: Eighteen prognostic MRGs (DNM1L, PUSL1, CHCHD4, COX7A1, CPT1A, CYP27A1, POLDIP2, PCK2, MRPL2, PDK3, PDK4, MARC2, ACSM3, COA7, THNSL1, ATAD3B, C15orf48, TOMM70A) were identified to construct the risk model. Remarkable discrepancies were observed between groups. The high-risk group had shorter overall survival, less immune infiltration, lower CD20 and higher PD-L1 expression than the low-risk group. Distinct immune microenvironment, responses to immunotherapy and predictive drug IC50 values were found between groups. CONCLUSIONS: We established a novel prognostic mitochondria-related signature by machine learning algorithm, which also demonstrated outstanding predictive value in tumor microenvironment and responses to therapies.