Machine learning methods for neuronal identification from electrophysiological recordings in the cerebellum
Prepared at the Wolfson Institute for Biomedical Research, University College London
Author: Gabriela Martinez
contact: airamgabriela17@gmail.com
Abstract: the cerebellum is a fundamental component of the vertebrate brain that is mainly responsible for physical coordination and movement learning. Its internal working mechanisms are still beyond understanding mainly due to the inability of expert neuroscientists to confidently classify observed cells in the cerebellar cortex based on their electrophysiological behavior, this is, the way in which neurons conduct electricity. Thanks to the advances in microchip technology and computational power, it is now possible to use machine learning to translate this biological problem into a multi-label classification one. In this study, neuron samples previously taken from the mouse cerebellum were processed and analyzed to feed a set of unsupervised and supervised methodologies with state-of-the-art modeling features that allowed the identification and prediction of four of the five types of cells that constitute the cerebellar cortex: Purkinje, Golgi, Granule, and Mossy Fibers, with precision levels that outperform a baseline model set according to experts’ assessment.
Key words: Cerebellar cortical cells, Cerebellar cortex, Cerebellum, High-precision classifier, Machine Learning, Neuronal prediction, Unsupervised clustering, Supervised classification