Fig. 5: Technical validation, transcriptomic and proteogenomic contextualization and data-driven immunohistochemical panel classifier.

a Exemplary immunoblot (11 samples out of the n = 54 in the technical validation sub cohort across five immunoblots) of ITGB1 (Integrin beta) and CFL1 (Cofilin-1); Std: standard; b Correlation of immunoblot and LC-MS/MS abundances of ITGB1; Spearman’s r; c Schematic overview of the classifier used for re-classification of the TCGA samples based on their transcriptomic data; d Mutational, structural variant, and epigenetic information from the reclassified TCGA dataset; top panel: customized OncoPrint of the 20 most frequent mutations; middle panel: corresponding relevant copy number variations; lower panels: further epigenetic and clinical data; small insets on the right visualize (per gene) the absolute frequency of mutations/amplifications/deep deletions (upper bar) and the relative frequencies per PAULA cluster (middle bar) or TCGA subtype (lower bar); e Kaplan-Meier overall survival plot of the validation dataset with all samples that could be reclassified with ≥50% cluster probability; p-value from log-rank test; f Upper panel: the top n = 9 proteins from the feature selection for the data-driven immunohistochemical (IHC) validation; red: insufficient correlation with LC-MS/MS; blue: final selection; grey: not evaluated; lower panel: in silico estimates (violinplot; boxplot: whiskers show 95%-interval, box is interquartile range, horizontal line is median, horizontal grey bar is mean) and observed values (asterisk) of accuracy and cluster I/III prediction error; g IHC and LC-MS/MS correlations of four exemplary proteins from the seven proteins investigated in subplot f; PPC: percentage of positive cells; h Exemplary micrographs, 25× objective; correlation coefficients r and p-values from Spearman’s rank correlation; i confusion matrix and Sankey plot of the IHC cluster predictions. Source data are provided as a Source Data file.