Multi-layer Bundling as a New Approach for Determining Multi-scale Correlations Within a High-Dimensional Dataset

by Mehran Fazli, Richard Bertram & Deborah Striegel

The multi-layer bundling (MLB) method delivers a robust new approach to cluster elements of complex biological networks. Using different partitioning schemes (clustering regimes) obtained by spectral clustering on the affinity matrix, MLB provides hierarchical layers of clusters called bundles, where each bundle in a layer is formed from all elements with the same membership throughout all partitioning schemes used up to the current layer. This iterative process offers profound insights into the interconnections among data elements not apparent through a single clustering approach. For example, MLB excels in identifying critical bridge sets within interacting systems. If removed, these bridges can disrupt system compartments and halt information propagation, making their identification crucial for understanding network integrity. Moreover, MLB's unique capability to integrate bundle membership information through multiple layers with the affinity matrix significantly enhances its predictive power in network reconstruction. Compared to methods like WGCNA, MLB offers a more robust and versatile approach. Requiring fewer user-defined parameters, MLB provides a clearer view of the underlying data structures, empowering researchers with a powerful tool to decode complex datasets and uncover meaningful biological insight. Its versatility extends beyond biological networks, making it valuable for various research domains.