Cross-disease stratification by genetic classifier Lymphly reveals shared molecular programs across B-cell malignancies
Authors
Dmitrii Snitkin, Pavel Zemskiy, Andrey Suponin, Mark Meerson, Ekaterina Nesterenko, Alexander Bagaev, Alexander Nesmelov, Konstantin Chernyshov, Nikita Kotlov
- BostonGene Corporation, Waltham, MA, USA
Abstract
Context:
Diffuse large B-cell lymphoma (DLBCL) comprises genetically distinct subtypes, but the molecular relationships between these subtypes and other B-cell lymphomas are poorly defined. Lymphly is an interpretable rule-based classifier that refines DLBCL genetic subtyping based on curated biologically informed genomic features.
Objective: Assess Lymphly’s ability to identify shared molecular programs across DLBCL and related B-cell lymphomas and to perform clinically relevant cross-disease stratification.
Design:
Integrative retrospective analysis of internal and publicly available molecular datasets.
Methods: Lymphly was applied to 4,353 DLBCL and 2,203 non-DLBCL B-cell lymphoma samples with available molecular data. Samples were assigned to canonical subtypes (EZB, MCD, BN2, N1), newly refined subtypes (JS3, JS6), and orthogonal genomic statuses. Cross-disease subtype mapping, concordance with established classifiers and survival for the DLBCL meta-cohort were assessed.
Results:
Lymphly revealed a preferential alignment between several B-cell lymphoma diagnoses with established DLBCL genetic subtypes, indicating shared molecular programs across related B-cell malignancies. Follicular lymphoma and Burkitt lymphoma samples were predominantly assigned to EZB, consistent with shared germinal center biology. Splenic marginal zone lymphoma was enriched for BN2, primary central nervous system lymphoma for MCD, primary mediastinal B-cell lymphoma for JS6, and T-cell/histiocyte-rich and plasmablastic lymphomas for JS3. Therefore, rather than forming isolated diagnosis-specific groups, these entities were distributed along shared axes of genetic subtypes, indicating convergent molecular programs. In an expanded DLBCL meta-cohort integrating whole-exome and targeted panel datasets, Lymphly was more robust than LymphGen in preserving subtype assignments, as evidenced by higher Jaccard concordance with full-feature classifications and fewer unclassified cases even in datasets with limited panel-derived genomic input. N1 and MCD were associated with the poorest overall survival, whereas BN2, EZB, JS3, and JS6 showed more favorable outcomes. Lymphly subtype assignment remained predictive of outcomes independently of the International Prognostic Index risk groups and differences between cohorts.
Conclusions:
Lymphly is applicable beyond DLBCL because it captures shared molecular architectures across related B-cell lymphomas and retains prognostic relevance across heterogeneous sequencing datasets, including targeted gene panels. Our findings support the use of this interpretable cross-disease genomic framework for disease stratification, translational studies, and biomarker-driven clinical trial design.
Diffuse large B-cell lymphoma (DLBCL) comprises genetically distinct subtypes, but the molecular relationships between these subtypes and other B-cell lymphomas are poorly defined. Lymphly is an interpretable rule-based classifier that refines DLBCL genetic subtyping based on curated biologically informed genomic features.
Objective: Assess Lymphly’s ability to identify shared molecular programs across DLBCL and related B-cell lymphomas and to perform clinically relevant cross-disease stratification.
Design:
Integrative retrospective analysis of internal and publicly available molecular datasets.
Methods: Lymphly was applied to 4,353 DLBCL and 2,203 non-DLBCL B-cell lymphoma samples with available molecular data. Samples were assigned to canonical subtypes (EZB, MCD, BN2, N1), newly refined subtypes (JS3, JS6), and orthogonal genomic statuses. Cross-disease subtype mapping, concordance with established classifiers and survival for the DLBCL meta-cohort were assessed.
Results:
Lymphly revealed a preferential alignment between several B-cell lymphoma diagnoses with established DLBCL genetic subtypes, indicating shared molecular programs across related B-cell malignancies. Follicular lymphoma and Burkitt lymphoma samples were predominantly assigned to EZB, consistent with shared germinal center biology. Splenic marginal zone lymphoma was enriched for BN2, primary central nervous system lymphoma for MCD, primary mediastinal B-cell lymphoma for JS6, and T-cell/histiocyte-rich and plasmablastic lymphomas for JS3. Therefore, rather than forming isolated diagnosis-specific groups, these entities were distributed along shared axes of genetic subtypes, indicating convergent molecular programs. In an expanded DLBCL meta-cohort integrating whole-exome and targeted panel datasets, Lymphly was more robust than LymphGen in preserving subtype assignments, as evidenced by higher Jaccard concordance with full-feature classifications and fewer unclassified cases even in datasets with limited panel-derived genomic input. N1 and MCD were associated with the poorest overall survival, whereas BN2, EZB, JS3, and JS6 showed more favorable outcomes. Lymphly subtype assignment remained predictive of outcomes independently of the International Prognostic Index risk groups and differences between cohorts.
Conclusions:
Lymphly is applicable beyond DLBCL because it captures shared molecular architectures across related B-cell lymphomas and retains prognostic relevance across heterogeneous sequencing datasets, including targeted gene panels. Our findings support the use of this interpretable cross-disease genomic framework for disease stratification, translational studies, and biomarker-driven clinical trial design.
Latest publications
