Fig. 2: Unsupervised clustering concordance across omics layers.

a–d Proportion of Ambiguity Clustering (PAC) values for transcriptome, proteome, metabolome, and phenome datasets across clustering algorithms (AP, BLOCK, CMEANS, DIANA_Euclidean, GMM, HC_Euclidean, KM, NMF_Brunet, NMF_Lee, PAM_Euclidean, SC, SOM) over cluster numbers k = 2 to 6. Lower PAC values indicate higher clustering stability. e–g Alluvial plots mapping patient transitions between omics-derived clusters. Sample distribution across clusters for metabolome, phenome, proteome, and transcriptome datasets at k = 2, 3, 4. Stacked bars reflect cluster assignments for each omics type, illustrating alignment of patient subgroups across transcriptome, proteome, metabolome, and phenome layers. AP affinity propagation, BLOCK biclustering using a latent block model, CMEANS fuzzy C-means clustering, DIANA_Euclidean dIvisive ANAlysis clustering, GMM Gaussian mixture model using Bayesian information criterion on EM algorithm, HC_Euclidean hierarchical clustering, KM K-means clustering, NMF nonnegative matrix factorization, PAM_Euclidean Partition Around Medoids, SC spectral clustering using Radial-Basis kernel function, SOM self-organizing map, PAC proportion of ambiguous clustering.