Fig. 5

Discovered survival signatures in DLBCL. a The top survival-associated GE signature was detected by SDCM in iteration k = 27. Depicted are log₂(ratios) for the top 40 genes by gene weights |v^g〉 (Eq. 21), ordered by their signature-specific strengths \(\left| {{\mathbf{u}}_{k,\hat o}^{\mathrm{g}}} \right\rangle\) (Eq. 29; green triangles in rows). Samples are ordered by their signature-specific sample strengths (light orange triangles in columns). Heatmaps show GE intensities in the detection cohort (left), the microarray-based validation cohort (center) and the RNA-sequencing based validation cohort (right). Colored lines below heatmaps indicate outcome (light red; PFS available for microarray-based cohorts; OS for RNA-sequencing) and subtype. In all cohorts, ABC DLBCL cases (light blue) are more frequent at negative sample strengths, while better outcome occurs for higher sample strengths. Signature details on transcript cluster level are presented in Supplementary Fig. 26. Gray lines indicate missing matches between platforms. Tabular definitions (signature axes and strengths for microarray-based cohorts) are provided in Supplementary Data 1 and 2. b The second survival-associated GE signature in the best predictor model was detected in iteration k = 11. As in (a), its top 40 genes are displayed. Similar to (a) and in all cohorts, colored lines indicate a higher frequency of GCB DLBCL cases (orange) at higher sample strengths. Other than in (a), these are paired with an untypically high rate of progressions or deaths for this subtype (light red). (Signature details on transcript cluster level in Supplementary Fig. 27; gray lines as in (a); tabular definitions in Supplementary Data 1 and 2).