Machine Learning Identifies Negative Emotions Encoded in the Cerebellum

This study investigates the cerebellum’s role in processing stress-induced negative emotions by analyzing neurophysiological signals from the deep cerebellar nuclei (DCN) in mice. We compared signals during chronic restraint stress and the tail suspension test, using implanted microwire electrodes to collect data from various DCN subnuclei, with a focus on the dentate and interstitial nuclei. Seventeen machine learning classifiers were employed to identify emotion encoding within low-frequency (0.5-49Hz) local field potentials (LFPs) in the cerebellar dentate nucleus. Notably, Medium Gaussian SVM, Medium Neural Network, Wide Neural Network, and Bilayered Neural Network demonstrated high accuracy in classifying emotional states via the cerebellar dentate nucleus, interstitial nucleus, and DCN. Our work proposes four classifiers suitable for distinguishing cerebellar negative emotion valence and provides evidence for the cerebellum’s role in emotion encoding from a machine learning perspective. This research offers new insights into the neural circuitry mechanisms of depression and theoretical support for developing novel neuromodulation paradigms to treat depression.