Machine learning-based classification of diffuse large B-cell lymphoma patients by eight gene expression profiles

Area of Science:

• Oncology
• Molecular Biology
• Bioinformatics

Background:

• Gene expression profiling (GEP) classifies diffuse large B-cell lymphoma (DLBCL) into GCB, ABC, and UC subtypes with prognostic significance.
• Current GEP classification is clinically impractical due to the large number of genes (>1000) and complex interpretation.

Purpose of the Study:

• To develop a simplified, clinically applicable model for stratifying DLBCL patients into GCB and Non-GCB subgroups.
• To validate the prognostic significance of the new model in patients treated with CHOP/R-CHOP chemotherapy.

Main Methods:

• Selected eight key genes (MYBL1, LMO2, BCL6, MME, IRF4, NFKBIZ, PDE4B, SLA) from Gene Expression Omnibus (GEO) data.
• Developed a support vector machine (SVM) model using receiver-operating characteristic (ROC) curves to estimate subgroup membership.
• Validated the model in independent cohorts totaling 855 DLBCL patients.

Main Results:

• The SVM model achieved high concordance (91.0%-94.4%) with GEP classification in training and validation cohorts.
• Patients classified as Non-GCB subtype exhibited significantly poorer outcomes compared to the GCB subtype.
• The model's prognostic power was independent of the International Prognostic Index (IPI).

Conclusions:

• A novel, simplified gene-based model effectively stratifies DLBCL patients into prognostically relevant GCB and Non-GCB subgroups.
• This model accurately predicts treatment response to CHOP/R-CHOP chemotherapy, offering a practical alternative to complex GEP.
• The findings support the clinical utility of this simplified approach for DLBCL patient management.