Supplementary Figure 2: Unsupervised analysis and cluster reliability for 212 GEP-NET samples.

a, Scatterplots showing the first five principal components, capturing 33% of the variance for 212 GEP-NET expression profiles. The tissue of origin is indicated by different colors. Primary tumors are shown with circles, while METs are shown with triangles. b, Two-dimensional tSNE projection for the expression data. Different colors indicate the different tissue of origin. c, Two-dimensional tSNE projection of the VIPER-inferred protein activity for 212 GEP-NET samples. The color of the symbols indicates tissue of origin, and their shape indicates status as primary tumors (circles) or METs (triangles). The color of the clouds indicates cluster membership according to Fig. 1b. d, Integrated reliability score for different cluster structures (different number of clusters) for the consensus cluster of 212 GEP-NET expression profiles (red) or VIPER-inferred protein activity profiles (blue). e, Probability density plot for cluster reliability estimated from the expression profiles and VIPER-inferred protein activity profiles for 212 GEP-NET samples (see g). f, Integrated reliability score for the complete cluster structure computed as the area over the cumulative probability curve. g, Cluster reliability score for 212 GEP-NET expression and VIPER-inferred protein activity profiles after consensus clustering in five clusters. The horizontal black line indicates the threshold for FDR < 0.01. h,i, Cluster reliability (h) and silhouette score (i) for each sample from the four-cluster structure based on expression and the five-cluster structure based on VIPER-inferred protein activity data. j, Cluster membership for the H-STS xenograft model. Shown is enrichment of the samples from each of the five clusters on the distance to the xenograft model based on the correlation between protein activity signatures. Enrichment significance is shown as –log₁₀ (P value) by the bar plot (one-tailed aREA test).