28/02/25
Exciting new work from an international collaboration between Dimitar Kostadinov and colleagues, published in Cell, introduces a new approach to identify neuronal cell types in high-density electrophysiological recordings from behaving animals.
Our brains generate behaviour through orchestrated activity across a variety of classes of neurons embedded in interconnected neural circuits. Understanding these computations requires deciphering the simultaneous action of these neural populations. The recent development of high-density probes has revolutionised our ability to record from numerous neurons simultaneously, but distinguishing between the different classes of neurons in a recording has remained a significant challenge. In this study, the team developed a semi-supervised deep learning classifier that can predict cell types with over 95% accuracy using the anatomical and functional features of each recorded neuron.
The researchers tested their strategy in the cerebellum, a region with well-defined neuronal connectivity and a variety of cell types. By combining optogenetics and pharmacology, they created a ground-truth library of properties for Purkinje cells, molecular layer interneurons, Golgi cells, and mossy fibres. This library was used to train the classifier, which the team then validated across different laboratories, probes, and species to demonstrate its robustness and generalisability. Applying this classifier in behaving mice and monkeys, the team then employed modern dynamic systems analysis to reveal the unique contributions of different cell types to cerebellar circuit computations during behaviour.
This work illustrates the exciting potential of integrating machine learning with neuroscience to tackle some of the most complex biological questions. The success of this approach in the cerebellum bodes well for further work and suggests the methodology could be adapted for other brain regions, allowing for comprehensive mapping of orchestrated neural circuit function in action. This significant advancement not only enhances our ability to understand physiological circuit function but also opens new routes to identify the impact of cell type-specific disruptions in a variety of brain disorders and to develop more targeted therapeutic interventions.
