MULTIFRACTAL CONNECTOME EIGENMODES LINK STRUCTURAL HETEROGENEITY TO MULTISCALE NEURAL REPRESENTATIONS

A long-standing goal in neuroscience is to explain how structural connectivity shapes brain function. Here, we leverage the synapse-resolved Drosophila connectome to uncover mechanistic links between network architecture and whole-brain activity. The fly connectome is highly heterogeneous, with heavy-tailed synaptic weights and neuronal in-degrees, as well as systematic correlations between connectivity and synaptic strength. We show that the eigenmodes of this network exhibit a unique organization, neither uniformly distributed (delocalized) nor confined to small regions (localized). Instead, they combine both localized and distributed features across multiple spatial scales, a characteristic described as multifractal. To identify the origin of this organization, we develop a generative network model that reproduces the key statistical features of the connectome and demonstrate that the joint heterogeneity of topology and synaptic strength is sufficient to give rise to multifractal eigenmodes. Using a connectome-constrained dynamical model, we further show that these modes support efficient, multiscale representations of sensory processing, coupling category-selective localisation with high-dimensional, distributed dynamics. We validate these predictions using whole-brain, single-neuron calcium imaging. Together, our results identify multifractal eigenmodes as an organizing principle that links heterogeneous connectome architecture to efficient neural coding, establishing a new structure–function relationship in the brain.