Classifier discovery. (A) Schematic diagram showing the strategy used to develop classifier on a public dataset (GSE10846) from newly diagnosed DLBCL patients (Discovery) and to classify new DLBCL specimens from an independent set of commercially sourced diagnostic/archival biopsies from frontline patients (Replication). Gene selection-steps are shown in pink, and classifier creation steps are shown in yellow. Heat maps showing the results of clustering of samples (Affymetrix) in the public dataset (n = 414) based on subgroup (B) and independent commercially sourced samples (NanoString) (n = 245) based on the classifier (C). Segregation according to the subgroup/classifier is shown in the lower bar above the heat map, whereas lack of segregation by COO is shown in the upper bar.