TOWARD A SPATIALLY AND PHYSIOLOGICALLY REALISTIC LARGE-SCALE MODEL OF THE CEREBELLAR NEURAL CIRCUIT

Large-scale, physiologically realistic computational models of neural circuits are becoming important tools to integrate multi-modal neuroscience data, from various physiological mechanisms to detailed connectome connectivity. The challenges of building such models lie not only in the large number of neurons and synapses but also in incorporating their distributions in 3D space, crucial for studying distributed computation in spatially extended neurons, such as the cerebellar Purkinje cells (PCs). Here we present our progress toward developing a large-scale, physiologically realistic model of the cerebellar neural circuit, building upon our previous efforts [1-3]. We employed container technology based on Singularity [4], enabling easy configuration and deployment of the model simulation built on CoreNEURON with GPUs [5]. In our preliminary results, this provided an efficient computational backend that can simulate a model with approximately one million granule cells in about five minutes per simulated second using four TITAN RTX GPUs. We also implemented the network of molecular layer interneurons (MLIs) based on recent connectome data revealing their disinhibitory connectivity [3,6]. Our simulations exhibited the spatiotemporal activation of MLIs induced by synchronous inputs that modulated PC dendritic activity. Therefore, our work underscores the importance of spatial and geometric information in neural circuits for understanding network-level information processing.
1. Sudhakar et al., PLOS Comp Biol, 2017
2. Kwon, Kim et al., Exp Mol Med, 2026 (in press)
3. Park et al., Nat Neurosci, 2026 (in press)
4. Kurtzer et al., PLoS ONE, 2017
5. Kumbhar et al., Front Neuroinf, 2019
6. Lackey et al., Neuron, 2024