MGN-Net: A multi-view graph normalizer for integrating heterogeneous biological network populations

With the recent technological advances, biological datasets, often represented by networks (i.e., graphs) of interacting entities, proliferate with unprecedented complexity and heterogeneity. Although modern network science opens new frontiers analyzing connectivity patterns in such we still lack data-driven methods for extracting an integral connectional fingerprint a multi-view graph population, let alone disentangling typical from atypical variations across population samples. We present normalizer (MGN-Net2), neural based method to normalize integrate set into single template that is centered, representative, topologically sound. demonstrate use MGN-Net discovering fingerprints healthy neurologically disordered brain populations including Alzheimer’s disease Autism spectrum disorder patients. Additionally, comparing learned templates populations, show significantly outperforms conventional integration extensive experiments terms producing most centered templates, recapitulating unique traits preserving complex topology networks. Our evaluations showed powerfully generic easily adaptable design different graph-based problems as identification relevant connections, normalization integration.