Coming up short: generative network models fail to accurately capture long-range connectivity

Stuart Oldham,Alex Fornito,Gareth Ball

DOI: https://doi.org/10.1101/2024.11.18.624192

2024-11-19

Abstract:Generative network models (GNMs) have been proposed to identify the mechanisms/constraints that shape the organisation of the connectome. These models most commonly parameterise the formation of inter-regional axonal connections using a trade-off between connection cost and some measure of topological complexity or functional value. Despite its simplicity, GNMs can generate synthetic networks that capture many topological properties of empirical brain networks. However, current models often fail to capture the spatial embedding-i.e., the topography-of many such properties, such as the anatomical location of network hubs. In this study, we investigate a diverse array of generative network model formulations and find that none can accurately capture empirical patterns of long-range connectivity. We demonstrate that the spatial embedding of long-range connections is critical in defining hub locations and that it is precisely these connections that are poorly captured by extant models. We further show how standard metrics used for model optimisation and evaluation fail to capture the true correspondence between synthetic and empirical brain networks. Overall, our findings demonstrate common failure modes of GNMs, identify why these models do not fully capture brain network organisation, and suggest ways the field can move forward to address these challenges.

Biology