Diffuse large B cell lymphoma (DLBCL) displays genetic and clinical heterogeneity that limits the predictive value of cell-of-origin and LymphGen. We evaluated whether diagnostic targeted next generation sequencing (NGS) could predict outcomes and delineate high-risk subsets in routine practice. We retrospectively analyzed tumors from 106 patients. Variants with allele frequency(VAF) ≥ 10% from a lymphoma-focused panel were curated into a binary mutation matrix. Random Forest regression identified genes associated with overall survival (OS) and progression-free survival (PFS). Principal-component analysis with k-means clustering stratified tumors, and groups were compared for survival, clinical and pathologic features, and LymphGen composition. Thirty genes with strong prognostic signals were recovered, spanning canonical drivers and novel candidates. Clustering of these features yielded two molecular subgroups, one highly enriched for LymphGen-unclassified tumors. This subgroup showed significantly inferior PFS and a higher prevalence of adverse clinical factors like advanced IPI, stage, elevated LDH. Targeted NGS combined with machine learning and unsupervised clustering provides robust, clinical significant stratification in DLBCL, revealing a high-risk subgroup not captured by existing classification. These findings support incorporating panel-based sequencing into routine diagnostics to refine risk-adapted treatment and prioritize patients for intensified or novel therapies.
Jang et al. (Mon,) studied this question.