Mapping Cancer Heterogeneity through Consensus Network Modeling: A Systems Biophysics Approach

Geng-Ming Hu^1*, Hsin-Wei Chen², Chi-Ming Chen¹

¹Physics, National Taiwan Normal University, Taipei, Taiwan
²Biological Science and Technology, National Yang Ming Chiao Tung University, Hsinchu, Taiwan

* Presenter:Geng-Ming Hu, email:pipipshu@gmail.com

Accurate cancer classification remains a fundamental challenge in molecular oncology. Here, we introduce consensus MSClustering, a multi-scale hierarchical network framework designed to integrate multi-omics data through an unsupervised consensus process. In this approach, each tumor sample is represented as a node within a high-dimensional expression similarity network, and a heterogeneity index is introduced to quantify system-level fluctuations. Applying this framework to 2,439 tumor samples across ten cancer types, we identified 167 functionally coherent gene modules that capture the underlying structure of molecular variability.

By combining the MSClustering algorithm with a consensus distance metric that integrates multiple omic platforms, the resulting network topology exhibits well-defined modular organization and hierarchical stability, enabling precise subtype delineation and uncovering conserved biological pathways across cancers. This integrative framework bridges computational biophysics and cancer systems biology, offering a quantitative perspective for characterizing disease heterogeneity from complex molecular expression signals.

Keywords: Unsupervised Learning, Consensus Clustering, Cancer Heterogeneity, Network Analysis