Rigid Graph Alignment

Abstract

An increasingly important class of networks is derived from physical systems that have a spatial basis. Specifically, nodes in the network have spatial coordinates associated with them, and conserved edges in two networks being aligned have correlated distance measures. An example of such a network is the human brain connectome – a network of co-activity of different regions of the brain, as observed in a functional MRI (fMRI). Here, the problem of identifying conserved patterns corresponds to the alignment of connectomes. In this context, one may structurally align the brains through co-registration to a common coordinate system. Alternately, one may align the networks, ignoring the structural basis of co-activity. In this paper, we formulate a novel problem – rigid graph alignment, which simultaneously aligns the network, as well as the underlying structure. We formally specify the problem and present a method based on expectation maximization, which alternately aligns the network and the structure via rigid body transformations. We demonstrate that our method significantly improves the quality of network alignment in synthetic graphs. We also apply rigid graph alignment to functional brain networks derived from 20 subjects drawn from the Human Connectome Project (HCP), and show over a two-fold increase in quality of alignment. Our results are broadly applicable to other applications and abstracted networks that can be embedded in metric spaces – e.g., through spectral embeddings.