Translating network models to parallel hardware in NEURON

doi:10.1016/j.jneumeth.2007.09.010

Journal of Neuroscience Methods

Volume 169, Issue 2, 30 April 2008, Pages 425-455

https://doi.org/10.1016/j.jneumeth.2007.09.010 Get rights and content

Abstract

The increasing complexity of network models poses a growing computational burden. At the same time, computational neuroscientists are finding it easier to access parallel hardware, such as multiprocessor personal computers, workstation clusters, and massively parallel supercomputers. The practical question is how to move a working network model from a single processor to parallel hardware. Here we show how to make this transition for models implemented with NEURON, in such a way that the final result will run and produce numerically identical results on either serial or parallel hardware. This allows users to develop and debug models on readily available local resources, then run their code without modification on a parallel supercomputer.

Section snippets

Uncited references

Bush et al. (1999), Carnevale and Hines (2006), Davison et al. (2003), Hines and Carnevale (2004), Kirkpatrick (2003), MCellRan4 (2007), Migliore et al. (2006), Santhakumar et al. (2005) and Sutter (2005).

References (9)

M.L. Hines et al.
Discrete event simulation in the NEURON environment
Neurocomputing
(2004)
P.C. Bush et al.
Increased pyramidal excitability and NMDA conductance can explain posttraumatic epileptogenesis without disinhibition: a model
J Neurophysiol
(1999)
N.T. Carnevale et al.
The NEURON book
(2006)
A.P. Davison et al.
Dendrodendritic inhibition and simulated odor responses in a detailed olfactory bulb network model
J Neurophysiol
(2003)

There are more references available in the full text version of this article.

Cited by (46)

Neural mechanisms of the temporal response of cortical neurons to intracortical microstimulation
2024, Brain Stimulation
Intracortical microstimulation (ICMS) is used to map neuronal circuitry in the brain and restore lost sensory function, including vision, hearing, and somatosensation. The temporal response of cortical neurons to single pulse ICMS is remarkably stereotyped and comprises short latency excitation followed by prolonged inhibition and, in some cases, rebound excitation. However, the neural origin of the different response components to ICMS are poorly understood, and the interactions between the three response components during trains of ICMS pulses remains unclear.
We used computational modeling to determine the mechanisms contributing to the temporal response to ICMS in model cortical neurons.
We implemented a biophysically based computational model of a cortical column comprising neurons with realistic morphology and synapses and quantified the temporal response of cortical neurons to different ICMS protocols. We characterized the temporal responses to single pulse ICMS across stimulation intensities and inhibitory (GABA-B/GABA-A) synaptic strengths. To probe interactions between response components, we quantified the response to paired pulse ICMS at different inter-pulse intervals and the response to short trains at different stimulation frequencies. Finally, we evaluated the performance of biomimetic ICMS trains in evoking sustained neural responses.
Single pulse ICMS evoked short latency excitation followed by a period of inhibition, but model neurons did not exhibit post-inhibitory excitation. The strength of short latency excitation increased and the duration of inhibition increased with increased stimulation amplitude. Prolonged inhibition resulted from both after-hyperpolarization currents and GABA-B synaptic transmission. During the paired pulse protocol, the strength of short latency excitation evoked by a test pulse decreased marginally compared to those evoked by a single pulse for interpulse intervals (IPI) < 100 m s. Further, the duration of inhibition evoked by the test pulse was prolonged compared to single pulse for IPIs <50 m s and was not predicted by linear superposition of individual inhibitory responses. For IPIs>50 m s, the duration of inhibition evoked by the test pulse was comparable to those evoked by a single pulse. Short ICMS trains evoked repetitive excitatory responses against a background of inhibition. However, the strength of the repetitive excitatory response declined during ICMS at higher frequencies. Further, the duration of inhibition at the cessation of ICMS at higher frequencies was prolonged compared to the duration following a single pulse. Biomimetic pulse trains evoked comparable neural response between the onset and offset phases despite the presence of stimulation induced inhibition.
The cortical column model replicated the short latency excitation and long-lasting inhibitory components of the stereotyped neural response documented in experimental studies of ICMS. Both cellular and synaptic mechanisms influenced the response components generated by ICMS. The non-linear interactions between response components resulted in dynamic ICMS-evoked neural activity and may play an important role in mediating the ICMS-induced precepts.
Stoney vs. Histed: Quantifying the spatial effects of intracortical microstimulation
2022, Brain Stimulation
Citation Excerpt :
The values of Ve were coupled to each neuronal compartment using the e_extracellular mechanism [36]. The simulation was parallelized by distributing the total number of neurons (36,000) across 400 processors in a round-robin fashion (90 neurons/processor) [47]. The following statistical tests were undertaken to infer effects of ICMS intensity on axonal and somatic activation distances.
Intracortical microstimulation (ICMS) is used to map neural circuits and restore lost sensory modalities such as vision, hearing, and somatosensation. The spatial effects of ICMS remain controversial: Stoney and colleagues proposed that the volume of somatic activation increased with stimulation intensity, while Histed et al., suggested activation density, but not somatic activation volume, increases with stimulation intensity.
We used computational modeling to quantify the spatial effects of ICMS intensity and unify the apparently paradoxical findings of Histed and Stoney.
We implemented a biophysically-based computational model of a cortical column comprising neurons with realistic morphology and representative synapses. We quantified the spatial effects of single pulses and short trains of ICMS, including the volume of activated neurons and the density of activated neurons as a function of stimulation intensity.
At all amplitudes, the dominant mode of somatic activation was by antidromic propagation to the soma following axonal activation, rather than via transsynaptic activation. There were no occurrences of direct activation of somata or dendrites. The volume over which antidromic action potentials were initiated grew with stimulation amplitude, while the volume of somatic activation increased marginally. However, the density of somatic activation within the activated volume increased with stimulation amplitude.
The results resolve the apparent paradox between Stoney and Histed's results by demonstrating that the volume over which action potentials are initiated grows with ICMS amplitude, consistent with Stoney. However, the volume occupied by the activated somata remains approximately constant, while the density of activated neurons within that volume increase, consistent with Histed.
The open EEGLAB portal Interface: High-Performance computing with EEGLAB
2021, NeuroImage
Citation Excerpt :
In doing so, NSG aims to lower or eliminate the administrative and technical barriers that currently make it difficult for investigators to use these resources. The NSG (nsgportal.org) gateway offers free computer time to neuroscientists on the XSEDE computer network, providing a ready means to use HPC resources for popular neuroscience tools, pipelines, data processing, and software including NEURON (Hines and Carnevale, 2008), GENESIS (Bower and Beeman, 2012), MOOSE (Ray et al., 2008), NEST (Gewaltig and Diesmann, 2007), Brian (Goodman and Brette, 2009) and PyNN (Davison et al., 2009), as well as software for analyzing and visualizing structural and functional brain imaging data, such as Freesurfer (Fischl, 2012) and MRtrix (Tournier et al., 2012). NSG leverages access to resources including the supercomputers Comet at the University of California San Diego and Stampede2 at University of Texas Austin, and the Jetstream cloud resource at the University of Indiana.
EEGLAB signal processing environment is currently the leading open-source software for processing electroencephalographic (EEG) data. The Neuroscience Gateway (NSG, nsgportal.org) is a web and API-based portal allowing users to easily run a variety of neuroscience-related software on high-performance computing (HPC) resources in the U.S. XSEDE network. We have reported recently (Delorme et al., 2019) on the Open EEGLAB Portal expansion of the free NSG services to allow the neuroscience community to build and run MATLAB pipelines using the EEGLAB tool environment. We are now releasing an EEGLAB plug-in, nsgportal, that interfaces EEGLAB with NSG directly from within EEGLAB running on MATLAB on any personal lab computer. The plug-in features a flexible MATLAB graphical user interface (GUI) that allows users to easily submit, interact with, and manage NSG jobs, and to retrieve and examine their results. Command line nsgportal tools supporting these GUI functionalities allow EEGLAB users and plug-in tool developers to build largely automated functions and workflows that include optional NSG job submission and processing. Here we present details on nsgportal implementation and documentation, provide user tutorials on example applications, and show sample test results comparing computation times using HPC versus laptop processing.
Simplified parametric models of the dielectric properties of brain and muscle tissue during electrical stimulation
2019, Medical Engineering and Physics
Parametric models are commonly used to describe the dispersive dielectric properties of biological tissues. While distinct regions of dispersion have been identified, the relative contribution of each during electrical stimulation is unknown. This study quantified the contribution of individual poles in parametric models of brain and muscle dielectric properties during electrical stimulation. The effect on the extracellular voltage waveform and threshold current for nerve stimulation of selectively removing subsets of poles from Cole-Cole and Debye models was examined. Errors were introduced when dispersions below 100 kHz were removed in both brain and muscle tissue. Poles below 1 kHz influenced the amplitude of the extracellular voltage waveform and the predicted minimum stimulation current. Poles between 1 kHz and 100 kHz influenced the waveform shape, with a minor effect on stimulus amplitude. The results confirm that low frequency dispersion in conductivity and permittivity can fundamentally influence the electric field and neural response during stimulation and provide insight into the relative contribution of the different dispersive regimes. Furthermore, they provide justification for for simplifying parametric models of dielectric properties through the removal of high frequency poles above 100 kHz which could improve the efficiency of time-domain solvers for simulations involving time-varying or aperiodic stimuli as may be required for certain closed-loop stimulation paradigms.
Single-Neuron and Network Computation in Realistic Models of the Cerebellar Cortex
2016, The Neuronal Codes of the Cerebellum
The cerebellum, due to its regular internal connectivity and the limited number of neuronal types, has inspired a series of modeling efforts. These vary in complexity and aim, ranging from single-neuron to network and system control models. Initially, theoretical models provided remarkable predictions on how the cerebellum might operate in the context of the sensorimotor control system, though at the expense of oversimplifications in neuron and network representation. Recently, two main advances have occurred, promoted by the availability of new physiological knowledge and by informatics technology. First, detailed biophysically grounded models of neurons and synapses have been developed and are now reaching a high level of sophistication, allowing a careful analysis of single-neuron computation. Major advances have been made with granule cell, Golgi cell, unipolar brush cell, Purkinje cell, deep cerebellar nucleus, and inferior olivary cell models. Moreover, dynamic synaptic models and networks of local microcircuits have been constructed. Second, it has been possible to transform these models into real-time versions, which have subsequently been applied to robotic simulations. Notably, real-time spiking networks of the cerebellum have been endowed with plasticity rules and embedded into robotic control systems. Thus, it is now becoming possible to translate the biophysical properties of neurons and synapses into network functions and to observe how these are exploited in the context of simulated behaviors. These observations are providing new insights into how single-neuron and connectivity properties are exploited by the cerebellar network operating in a closed loop.
Reproducibility of biophysical in silico neuron states and spikes from event-based partial histories
2023, PLoS Computational Biology

View all citing articles on Scopus

View full text

Translating network models to parallel hardware in NEURON

Abstract

Section snippets

Uncited references

Neurocomputing

Increased pyramidal excitability and NMDA conductance can explain posttraumatic epileptogenesis without disinhibition: a model

J Neurophysiol

The NEURON book

Dendrodendritic inhibition and simulated odor responses in a detailed olfactory bulb network model

J Neurophysiol