Decoding the Brain: Deep Learning Identifies Cell Types Across Species

In a major advance for systems neuroscience, researchers developed a deep-learning strategy to identify neuron cell types from high-density extracellular recordings in awake animals—without invasive labeling. Using the cerebellum as a test system, the team built a ground-truth dataset via optogenetics and pharmacology to classify Purkinje cells, interneurons, Golgi cells, and mossy fibers. Their semi-supervised model, trained on waveform features, firing statistics, and recording depth, achieved over 95% accuracy and generalized robustly across probes, labs, cerebellar regions, and even species. This tool not only overcomes a long-standing barrier in interpreting electrophysiological data but also enhances our ability to map and understand functional contributions of specific cell types during behavior.