Extended Data Fig. 5: NMF consensus-clustering identifies three proteomic subtypes of HBV-related early-stage HCC. | Nature

Extended Data Fig. 5: NMF consensus-clustering identifies three proteomic subtypes of HBV-related early-stage HCC.

From: Proteomics identifies new therapeutic targets of early-stage hepatocellular carcinoma

Extended Data Fig. 5

a, Principal component analysis; first two principal components of protein intensities, with samples connected by centroids according to sample types (green, non-tumour; red, tumour). The tumour samples (red, n = 101) exhibit higher heterogeneity than the non-tumour (green, n = 98) samples. The ellipse presents the 0.9 confidence intervals for each type. b, c, Heat map of NMF consensus matrix, cophenetic correlation coefficient and average silhouette-width plots. The NMF input is the quantile-normalized iBAQ intensity matrix of the top 25% most-variant proteins, across 101 tumour samples. On the basis of visual inspection of the hierarchical clustering and the profiles of cophenetic correlation coefficient and average silhouette width for solutions with 2 to 7 clusters, we consider K = 3 to be the preferred solution (indicated by black triangles) and use this to arrange the samples shown in Fig. 2a (yielding the three clusters highlighted in blue, yellow and red). Average cophenetic correlations of each rank of clusters are displayed under the heat map matrix. The green and blue points or lines in the cophenetic correlation and average silhouette-width plots represent the non-smooth NMF (nsNMF) and the Brunet algorithm, respectively.

Source data