Connectome analysis of a cerebellum-like circuit for sensory prediction

Stable and accurate perception involves comparing incoming sensory input with internally- generated predictions ^1–3 . A mechanistic understanding of this process has been elusive due to the size and complexity of the relevant brain regions in mammals. Here we leverage connectomics to comprehensively map the cell types and synaptic connections underlying a well-characterized and ecologically relevant form of predictive sensory processing in the cerebellum-like electrosensory lobe (ELL) of weakly electric fish ^4,5 . Connectome analysis reveals highly-structured feedforward and recurrent synaptic connectivity mediating the cancellation of predictable electrosensory input. A computational model constrained by prior electrophysiological recordings shows how this connectivity supports the formation of predictions at multiple sites within the network and how the ELL solves a continual learning problem by maintaining fast and accurate predictions despite noise and changes in environmental context. Overall, these findings provide a blueprint for using connectomics to elucidate learning in vertebrate nervous systems.