Researchers set out to simplify and broaden molecular classification in DLBCL by moving from DNA-only algorithms to a transcriptional framework. They built the platform, called LGsig, using paired genomic and transcriptomic data from N = 764 cases drawn from the National Cancer Institute/British Columbia Cancer Agency cohort. The development cohort was restricted to tumors that had been assigned to established LymphGen (LG) classes defined by specific genetic lesions: MYD88 L265P and CD79B mutations (MCD); BCL6 translocation and NOTCH2 mutations (BN2); EZH2 mutations and BCL2 translocation (EZB); and SGK1/TET2 mutations (ST2).
Gene features were selected by differential expression. The authors identified an optimal panel of 294 genes and trained a nearest shrunken centroid classifier. The classifier yields four transcriptional analogues of the LG genetic categories, labeled MCDsig, BN2sig, ST2sig, and EZBsig. The analytical choice prioritized a compact gene set amenable to RNA-based classification while preserving subtype-specific transcriptional signals. The abstract reports that LGsig accurately identifies LG subtypes using RNA alone, but the abstract did not report classifier sensitivity, specificity, or confusion-matrix details; those metrics may be present in the full text.
The study team observed that many DLBCLs left unassigned by genetic-only algorithms could be placed into biologically coherent groups by transcriptional profiling. LGsig extended subtype assignments to all previously LG-unclassified cases. According to the authors, reassigned tumors showed transcriptional and clinical features mirroring their LG counterparts—meaning the expression-based labels aligned with the expected biology and outcomes of the corresponding genetically defined groups.
A particular reclassification noted in the abstract concerns the LG A53 group. LG A53 tumors are characterized in prior work by aneuploidy and TP53 alterations. The investigators reported that LGsig redistributed A53 cases into the four LGsig clusters in ways that the authors considered more biologically and therapeutically relevant than leaving them as a separate A53 genetic category. The abstract does not provide the proportions of A53 reassignment or downstream outcome metrics for those redistributed cases; the manuscript should be consulted for those granular data.
To test clinical utility, the authors applied LGsig to archival RNAseq from the POLARIX randomized trial (N = 678). POLARIX compared polatuzumab vedotin plus R-CHP (pola-R-CHP) with standard R-CHOP. Using LGsig classifications, the study team assessed outcomes by transcriptional subtype. The abstract states that LGsig improved the biomarker performance of LG within POLARIX by identifying distinct DLBCL subtypes that exhibited a survival benefit with pola-R-CHP over R-CHOP. This signal was observed in both cases originally classified by LG and those that had been LG-unclassified and subsequently reassigned by LGsig.
The abstract does not list numerical effect sizes, hazard ratios, confidence intervals, or p-values for the reported survival benefit by subtype. It also does not specify which LGsig subtypes drove the observed pola-R-CHP advantage. Those outcome details and subgroup statistics are not included in the abstract and would require consultation of the full article for precise quantitative results.
Authors reported accurate subtype identification from transcriptional data and extension of assignments to previously unclassified tumors. They used a nearest shrunken centroid classifier built on 294 genes. The abstract claims that transcriptionally reassigned tumors mirrored both transcriptional and clinical features of their genetically defined counterparts, supporting the face validity of the approach.
The abstract does not include the technical performance metrics typically used to evaluate classifiers—such as cross-validation accuracy, area under the ROC curve, or external validation statistics—so readers should look to the full text for these details. The applied dataset from POLARIX (N = 678) functioned as a clinically relevant validation cohort, enabling assessment of therapeutic associations, though numeric validation outputs are not provided in the abstract.
The authors conclude that LGsig expands molecular classification beyond current genetic classifiers by integrating genetic and transcriptional dimensions to better inform subtype-specific therapeutic strategies. The abstract explicitly stops short of providing granular statistical outcomes in the summary, and it does not report classifier performance metrics or exact effect sizes for POLARIX subgroups. The abstract also does not describe implementation logistics—such as assay platform, sample requirements, turnaround time, or how to operationalize LGsig in routine practice.
Key unanswered questions that remain (based on the abstract): which specific LGsig subtypes showed the pola-R-CHP survival benefit; what were the magnitude and statistical certainty of those differences; how reproducible are LGsig calls across RNA platforms and sample types; and whether prospective trials using LGsig-guided therapy selection have been planned. These items are not reported in the abstract and require inspection of the full manuscript or subsequent communications for confirmation.