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

摘要

Gene expression profiling (GEP) had divided the diffuse large B‐cell lymphoma (DLBCL) into molecular subgroups: germinal center B‐cell like (GCB), activated B‐cell like (ABC), and unclassified (UC) subtype. However, this classification with prognostic significance was not applied into clinical practice since there were more than 1000 genes to detect and interpreting was difficult. To classify cancer samples validly, eight significant genes (MYBL1, LMO2, BCL6, MME, IRF4, NFKBIZ, PDE4B, and SLA) were selected in 414 patients treated with CHOP/R‐CHOP chemotherapy from Gene Expression Omnibus (GEO) data sets. Cutoffs for each gene were obtained using receiver–operating characteristic curves ( style="fixed-case">ROC) new model based on the support vector machine ( style="fixed-case">SVM) estimated the probability of membership into one of two subgroups: style="fixed-case">GCB and Non‐ style="fixed-case">GCB ( style="fixed-case">ABC and style="fixed-case">UC). Furtherly, multivariate analysis validated the model in another two cohorts including 855 cases in all. As a result, patients in the training and validated cohorts were stratified into two subgroups with 94.0%, 91.0%, and 94.4% concordance with style="fixed-case">GEP, respectively. Patients with Non‐ style="fixed-case">GCB subtype had significantly poorer outcomes than that with style="fixed-case">GCB subtype, which agreed with the prognostic power of style="fixed-case">GEP classification. Moreover, the similar prognosis received in the low (0–2) and high (3–5) style="fixed-case">IPI scores group demonstrated that the new model was independent of style="fixed-case">IPI as well as style="fixed-case">GEP method. In conclusion, our new model could stratify style="fixed-case">DLBCL patients with style="fixed-case">CHOP/R‐ style="fixed-case">CHOP regimen matching style="fixed-case">GEP subtypes effectively.