Machine Learning‐Based Stem Cell‐Like Phenotype Identification and Novel Risk Stratification in Diffuse Large B‐Cell Lymphoma: Multi‐Omics Data from Multicenter Studies

Background: Previous researches have mainly focused on the discussion of whether cancer stem cells exist in diffuse large B-cell lymphoma (DLBCL). However, some subgroups with dismal prognosis that exhibit cell-like characteristics been overlooked. Methods: Using an innovative one-class logistic regression (OCLR) machine learning algorithm, we calculated stemness indices and systematically assessed their ability to reflect oncogenic dedifferentiation-like DLBCL. Next, identified signatures associated features using LASSO SVM-RFE algorithms developed a proportional hazards model (Riskscore) for DLBCL through Cox regression. Then stratified risk (n = 2133) evaluated prognostic value across known clinical molecular subgroups. We further compared high- low-risk Burkitt lymphoma. Finally, investigated mechanisms poor high-risk transcriptomics, genomics 576) single-cell RNA sequencing 19) data, as well cell experiments internal validation cohorts. Results: In this study, validated subgroup (25.6% DLBCL) prognosis. This group was defined polyamine metabolism-cold immune tumor upregulated metabolic key enzyme (ODC1) desert-like infiltration, had lower 3-year OS rate (54.3% vs. 83.6%, p < 0.0001) PFS (42.8% 75.2%, group. found patients MYC rearrangement, double-hit, double-expresser, or complete remission may either favorable prognosis, which can be accurately by our stratification model. Additionally, exhibited malignant proliferative phenotypes similar Genomic analysis revealed widespread copy number losses chemokine interferon coding regions 8p23.1 9p21.3 Bulk transcriptome indicated upregulation ODC1 might mediate cold microenvironment DLBCL, knocking down effectively inhibited proliferation. Promisingly, who were insensitive immunotherapy (CAR-T checkpoint antibodies). The research funded by: work supported grants from National Natural Science Foundation China (grant no. 81873450, 82170181), Beijing Hospitals Authority Youth Programme (code: QML20200201), (No. 7222027) Liang Wang. Keywords: Bioinformatics, Computational Systems Biology, Microenvironment, Risk Models No conflicts interests pertinent abstract.