Efficient in silico 3D intracellular neuron simulation

TitleEfficient in silico 3D intracellular neuron simulation
Publication TypeConference Paper
Year of Publication2019
AuthorsNewton, A.. J. H., Conte C.., Eggleston L.., Blasy E.., Hines M.. L., Lytton W.. W., & Mcdougal R.. A.
Conference NameSociety for Neuroscience 2019 (SFN '19)
Keywords2019, sfn 2019, Society for Neuroscience
Abstract

The activity patterns of neurons are governed by nonlinear interactions of their internal state and synaptic inputs; these interactions can be predicted using quantitative models. Since NEURON 7.4, the NEURON simulator has supported models coupling 3d intracellular – necessary for understanding microdomains near spines or the morphology variations where dendrites meet the soma – and electrophysiological kinetics, however computational overhead imposed practical limits on models that can be studied this way. The recently released NEURON 7.7 features completely redesigned 3D voxelization and parallel simulation algorithms, drastically reducing the compute time – expanding the set of simulatable models – without requiring changes to model implementation code. We are able to run a 4-thread 300ms simulation of a propagating wave on approximately 750k voxels near the soma of a virtual neuron with a reconstructed morphology in approximately 258s. Conversion of point-diameter neuron traces to 3D volumes follows the CTNG heuristic (McDougal et al., 2013) but extended so that discretization time scales with the volume of the neuron not the volume of the bounding box. Simulation uses an adaption of the parallel Douglas-Gunn algorithm (Newton et al., 2018) that runs on irregular domains. Reactions are JIT compiled and are exportable to SBML. Currents always enter and are computed based on surface voxels. We demonstrate scaling and simulation using models of the circadian oscillator, propagating calcium waves at the soma, and diffusion near spines.