Predicting Neural Activity from Connectome Embedding Spaces

While cortical connectomes contain enormous amounts of information, population activity is typically low-dimensional. Can low-dimensional connectome features reliably predict neural activity? Using the MICrONS dataset, which combines millimeter-scale, nanometer-resolution connectivity with simultaneously recorded in vivo activity, we find statistically significant alignment between morphological and functional similarity, quantified by subspace angles and centered kernel alignment. Topological analyses further show that representation spaces for both connectome and activity exhibit low-dimensional hyperbolic geometry with exponential scaling. Motivated by this shared geometry, we used multidimensional scaling to embed anatomical affinities and trained a simple linear model to reconstruct neuronal activity. The embedded connectome accounted for 68% of the variance in activity similarity, outperforming similarly simple models using the full high-dimensional connectome (56%). These results reveal robust structure-function coupling: geometry-aware dimensionality reduction discards most microscopic connectome detail yet improves prediction of neural activity. This suggests that synaptic wiring implicitly encodes an abstract low-dimensional organization underlying cortical population dynamics.