AI Neuron Synthesis Accelerates Brain Mapping

Google Research has created a synthetic neuron generation model called MoGen (Neuronal Morphology Generation) to improve neural connection reconstruction. The MoGen model, based on PointInfinity point cloud flow matching, generates realistic 3D neuron shapes using previously human-verified samples from the mouse cortex. When 10% synthetic data was added to the training set of the PATHFINDER model, the reconstruction error rate dropped by 4.4 percent, primarily due to a reduction in neurite merging errors. Although the improvement seems modest, at the scale of the full mouse brain map, this is equivalent to saving 157 person-years of manual verification. MoGen can generate specific neuron types by adjusting length, spatial extent, and branching. The researchers also trained versions of MoGen on data from the zebra finch and fruit fly. The model is released as open-source along with trained models for different species. The work will be presented at the ICLR 2026 conference.