Poster Z18 - REDEFINING BRAIN NETWORK COMPARISON WITH A NOVEL MULTISCALE DISTANCE MEASURE

Introduction: In 2021, neurological disorders ranked as the leading cause of disability-adjusted life-years (DALYs), affecting 3·40 billion individuals. Functional connectivity brain networks have demonstrated their ability to capture valuable insights about neurological disorders. However, no comprehensive network distance metric is available for comparing these brain networks. Here, we proposed a multiscale distance measure that considers the node and edge identities of corresponding brain areas. At the low scale, information concerning individual nodes and edges is captured. The mesoscale captures topological information driven by interactions between nodes. The high-scale information captures features of the network as an integrated entity.

Materials and

Methods: Our distance measure consists of three components, each corresponding to a distinct scale of information. In all equations, P and Q are the two networks of interest, D_i is the distance measure at the scale of interest, and JSD is the Jensen–Shannon divergence comparison between the distributions. For the low-scale component, the distribution of edge weights is compared between the two networks (Eq. (1) in Fig. 2). The mesoscale component utilizes the shortest paths between all pairs of nodes in the networks. Nodes that are one step, two steps, and up to N-steps away from each other are identified. The distribution of average weights for each N-step away neighborhood is compared using JSD to find the distances for the N-step away variation. The average distance of the M-1 neighborhoods is then computed (Eq. (2) in Fig. 2). The high-scale component compares the spectral representations of the two graphs by comparing eigenvalues and the distributions of eigenvectors of their Laplacian matrices (Eq. (3) in Fig. 2). In this equation, λ stands for eigenvalue, and v is the eigenvector.

Results, Conclusions, and Discussions: Our simulation analyses illustrated that each component of our distance measure can capture specific types of information not captured by the other measures, highlighting the multilevel storage of information in brain networks.
In this study, we proposed a new distance measure for comparing brain networks that captures information at multiple levels and considers both node and edge identity, enabling a more accurate estimation of dissimilarity between brain networks.

Acknowledgements (Optional):