Performance of the Lymph3Cx assay in the validation cohort and inter-laboratory concordance. (A) Heatmap of gene expression in the independent validation cohort of 88 pathologically defined PMBCL cases and 70 DLBCL cases. Each column represents an individual case, and the discriminating gene features included in the Lymph3Cx assay are ordered in rows. The top 6 genes are overexpressed in DLBCL, and the other 24 show higher expression levels in PMBCL. Housekeeping genes and the features used for DLBCL COO assignment are not displayed for visualization purposes. Cases are ordered according to their respective model score, with cases that obtained the highest score on the right. (B) Displayed are only the pathologically defined PMBCL cases with their respective assignment by the molecular gene expression–based assay (Lymph3Cx). No bias was observed with regards to mediastinal/nonmediastinal location of biopsy site, providing evidence that the assay is also applicable to PMBCL(-like) cases arising outside the mediastinum or nonmediastinal biopsy specimens, respectively. Interestingly, the 4 cases with a molecular signature of DLBCL (box) did not harbor chromosomal rearrangements of CIITA and/or the programmed death ligand (PDL) loci, or PDL copy-number (CN) alterations. (C) Distribution of model scores in the validation cohort. (D) Comparison of Lymph3Cx scores for selected cases of the validation cohort from 2 independent laboratories (BC Cancer Agency [BCCA] and Mayo Clinic). Dotted lines represent the defined thresholds to discriminate PMBCL (dark blue) from DLBCL (brown) using the Lymph3Cx assay (uncertain category displayed in purple). Of note, no case changed subtype assignment between the different laboratories. amp, amplification; ba, break-apart; CNA, copy-number alteration; N/A, not assessable.