Supplementary Figure 1: Analysis of inter-sample variation in the GEP-NET expression dataset and reliability analysis for different interactomes as models for GEP-NET. | Nature Genetics

Supplementary Figure 1: Analysis of inter-sample variation in the GEP-NET expression dataset and reliability analysis for different interactomes as models for GEP-NET.

From: A precision oncology approach to the pharmacological targeting of mechanistic dependencies in neuroendocrine tumors

Supplementary Figure 1

a, Violin plot showing the probability density for the distribution of mean squared error (MSE) computed between all closest sample pairs or the GEP-NET and 33 tumor types profiled by TCGA. The median value is indicated by a horizontal line. The number of samples for each tumor type is indicated on top of the figure. b, Bar plot showing the integrated network score computed as the area over the |NES| cumulative probability (Supplementary Fig. 3b,c). NES was computed by VIPER for 212 GEP-NET samples and all the regulatory proteins represented in the 29 evaluated interactomes (indicated inside the bars; Supplementary Table 3). When the network model is not representative of tissue-specific regulation, the master regulator analysis produces very few and barely significant results10. Here our GEP-NET interactome produced the strongest enrichment for 212 GEP-NET signatures when compared to 28 additional interactomes (Supplementary Table 3), indicating that GEP-NET is the best interactome, among all 29 tested ones, as a model for GEP-NET context-specific transcriptional regulation. c,d, Probability density for the functional conservation score of each regulon, expressed as z score (null model standard deviation units), between the GEP-NET interactome and interactomes assembled based only on metastases (MET; n = 4,340), primary tumors (Primary; n = 4,711), pancreas NETs (P-NET; n = 4,621) and small intestine NETs (SI-NET; n = 4,254) samples (shown by the filled histograms). Conservation of protein activity signatures computed from two disjoint subsets of each regulon (empty histograms) are shown as a reference point for the maximum achievable scores and as an indication of regulon robustness. The regulon functional conservation score was computed as the correlation between the VIPER-inferred activity signature for each protein across all NET tumors, as inferred from the GEP-NET interactome regulon, and regulons were assembled from specific subsets of samples, as previously described10. 99%, 98.9%, 97.2% and 94.5% of the GEP-NET interactome regulons were significantly conserved (FDR < 0.05) in the MET-, primary tumors–, P-NET- and SI-NET-specific interactomes, respectively, while the distribution for the functional conservation scores closely followed that of the maximal achievable conservation.