Benchmark of clustering performance in the ovarian cancer subtyping case study.

3D visualization revealed patient samples classified to different subtypes were much more tightly distributed in (a) the three-dimensional feature space based on ELM-CC clustering result (average Silhouette width = 0.37) than (b) in the space of top three principal components of gene expression profiles based on Consensus clustering result (average Silhouette width = 0.07). SigClust analysis showed more statistically significant (P < 0.05) pairwise comparisons of subtypes identified based on (c) ELM-CC than (d) the counterpart based on classical consensus clustering method.