From genetic stratification to multimodal stratification: precision diagnosis and treatment strategies for lymphoma

Diffuse large B-cell lymphoma (DLBCL) is a highly heterogeneous aggressive non-Hodgkin lymphoma. Although the R-CHOP regimen has served as the standard first-line therapy and substantially improved overall survival, its efficacy remains limited in biologically high-risk and treatment-resistant patients, highlighting the intrinsic limitations of uniform treatment strategies. Advances in high-throughput sequencing, single-cell technologies, and spatial profiling have established DLBCL as a collection of distinct molecular subtypes, each characterized by unique pathogenic mechanisms, therapeutic sensitivities, and patterns of resistance evolution. These discoveries have established the foundation for molecular subtype–guided precision therapy.Against this background, the Ruijin Hospital team proposed and implemented the GUIDANCE concept, a molecular classification–guided strategy for targeted therapy combined with chemotherapy. By developing a functional molecular classification system tailored to the Chinese population (LymphPlex) and prospectively matching molecular subtypes with specific targeted agents, the GUIDANCE-01 study significantly improved complete response rates and progression-free survival in newly diagnosed high-risk DLBCL. The GUIDANCE-06 study further demonstrated that this approach could enhance response depth and transplant eligibility in relapsed or refractory DLBCL, while the GUIDANCE-03 study successfully extended the strategy to peripheral T-cell lymphoma, underscoring its cross-lineage applicability. Collectively, these studies establish molecular classification not merely as a biological taxonomy, but as a clinically actionable decision-making framework.Nevertheless, molecular classification represents a static snapshot at a single time point and cannot fully capture dynamic tumor clonal evolution, immune microenvironment remodeling, or treatment-induced selective pressures. To address these limitations, recent efforts have focused on integrating multimodal data—including genomics, transcriptomics, radiomics, tumor microenvironment features, and longitudinal circulating tumor DNA monitoring—into artificial intelligence (AI)–based prognostic and stratification models. Such multimodal AI systems hold the potential to deliver continuous risk assessment and dynamic patient stratification, overcoming the incomplete coverage of conventional molecular subtyping.Looking forward, molecular classification is unlikely to be replaced by AI, but rather will serve as a foundational module within an AI-enabled precision oncology ecosystem. By combining molecular insights with real-time data integration and adaptive decision support, lymphoma management may transition from subtype-guided initial treatment toward comprehensive, longitudinal precision care.