MoBC-Net: computing network inter-module distance and module-based betweenness centrality

Analyzing disease-associated modules within gene co-expression and protein-protein interaction networks can improve disease-mechanism interpretation from multi-omics data. Disease-relevant interactions may also extend across module boundaries through genes not identified as differentially expressed, requiring consideration of both local module structure and the broader network context. Global centrality measures may preferentially rank highly connected hubs without distinguishing module-pair-specific connectors, while common size- and degree-matched random-module schemes do not explicitly account for local module compactness.

To address these challenges, we introduce MoBC‑Net, a multi‑scale network analysis framework that quantifies interactions among disease-associated modules and prioritizes candidate link genes. MoBC-Net selects interacting module pairs using a closest-distance measure, and identifies link genes via module betweenness centrality. Statistical significance is assessed using an adaptive binning and randomization scheme that preserves module size and node degree, while favoring local compactness. When applying MoBC-Net to TCGA breast cancer and cancer cachexia data, inter-module distances showed data-dependent concordance with those computed via gene ontology semantic similarity. In addition, link genes significantly overlapped in pathways commonly enriched by the module pair. Our case studies highlighted EPHA2 linking PI3K‑Akt modules in breast cancer and Gnai2 linking Apelin and cAMP signaling pathways affecting muscle hypertrophy and regeneration in cachexia. We believe that MoBC-Net's multi-scale approach, which progressively infers disease gene modules and key link genes, can improve network-guided investigation of disease mechanisms and therapeutic hypothesis generation.

An open-source R implementation of the MoBC-Net is available at GitHub (https://github.com/mogam-ai/MoBC-Net).