Identification of 2 subgroups with different immune infiltration patterns in a public dataset from newly diagnosed DLBCL patients. (A) Gene expression profiles (Affymetrix) of screening biopsies from 414 DLBCL patients downloaded from the Gene Expression Omnibus (GSE10846)⁷  were transformed by nonnegative matrix factorization, followed by unsupervised clustering, revealing 2 subgroups. The heat map (left panel) shows expression of top discriminating genes (n = 260) between the 2 subgroups that are independent of COO classification. The bands at the top show COO classification in comparison with new groups identified. PCA shows orthogonality between the 2 subgroups and COO classification (right panel). Principal component 1 segregates the 2 new subgroups identified, whereas principal component 2 shows segregation between ABC and GCB COO. Boxplots at the top and on the right show sample distribution by each of the 2 new subgroups identified and by COO for principal component 1 and principal component 2. (B) Results of GSEA pathway analysis presented using EnrichmentMap²³  to display the most significant biological processes (adj. P < .001) in subgroup A (left panel) and subgroup B (right panel). (C) Deconvolution plot of inferred immune cell populations shown as log₂-fold change in subgroup B relative to subgroup A. (D) Subgroup inclusion does not correlate with PFS in frontline DLBCL patients treated with R-CHOP from the GSE10846 dataset.