The NEURON Simulation Environment

A tool for empirically-based models of neurons and neural circuits

Active development and user support

Used by experimentalists, theoreticians, and educators for neuroscience research and teaching

As of March 2010� more than 900 publications� more than 1300 subscribers to Forum

and mailing list

The NEURON Simulation Environment

Open source project directed by Michael Hines

Active development and user support

Documentation, tutorials, and forum at www.neuron.yale.edu

CoursesSFN meetingssummer course at UCSDother courses

Source code for > 260 published models at www.modeldb.yale.edu

Hierarchies of Complexity

Structure

Single compartment

Stylized

Network

Anatomically detailed

v

Hierarchies of Complexity

Passive and Active currentsHH-stylekinetic scheme

Synaptic transmissioncontinuousspike-triggered

Gap junctions

Extracellular fields, Linear circuits

Diffusion, buffers, transport & exchange

Artificial spiking cells ("integrate & fire")

Mechanism

From Physical Systemto Computational Model

Conceptual model

a simplified representation of the physical system

Computational model

an accurate representation of the conceptual model

ComputationalModel

ConceptualModel

PhysicalSystem

From Physical Systemto Computational Model

dendrite

soma

Conceptualmodel

ballandstick

Physicalsystem

Ca1pyramidal

cell

Computationalmodel

hoccode

create soma, dendriteconnect dendrite(0), soma(1)v

Fundamental Concepts in NEURON

Signals

Electrical

Chemical

Whatmoves

chargecarriers

solute

Drivingforce

voltagegradient

concentrationgradient

What isconserved

charge

mass

Conservation of Charge

Cm

d Vm

d t� i

ion=� i

a

im

imim

im

ia

ia

ia

ia

The Model Equations

cj

dvj

d t� i

ionj

=�k

vk� v

j

r j k

vj

membrane potential in compartment j

iion

j

net transmembrane ionic current in compartment j

cj

membrane capacitance of compartment j

rjk

axial resistance between the centers ofcompartment j

andadjacent compartment k

Separating Anatomy and Biophysicsfrom Purely Numerical Issues

section

a continuous length of unbranched cable

Anatomical data from A.I. Gulyás

create soma, dendconnect dend(0), soma(1)

soma { L = 50 // [um] length diam = 50 // [um] diameter insert hh // Hodgkin-Huxley mechanism nseg = 1}

dend { L = 200 diam = 2 insert pas // passive channels nseg = 3}

Range Variables

Name Meaning Units

diam diameter [µm]

cm specific membrane [µf/cm2]

capacitance

g_pas specific conductance [siemens/cm2]

of the pas mechanism

v membrane potential [mV]

range

normalized position along the length of a section

0 � range � 1

any variable name can be used for range, e.g. x

0 1distance

normalized

0distancephysical

lengthphysical

nseg

the number of points in a section where

membrane current and potential are computed

nseg=1

nseg=2

nseg=3

Example: axon nseg = 3

To test spatial resolutionforall nseg = nseg*3

and repeat the simulation

Interpreters: hoc, Python

GUI (works via hoc)Import3DCellBuilder, Network Builder, Linear Circuit Builder

NMODL (compiled)ODEs, kinetic schemes, algebraic equations

accumulation, diffusion, transportreactionsion channels

events and state machines, artificial spiking cells

Model specification in NEURONModel specification in NEURON

// topology

create soma, dendconnect dend(0), soma(1)

// geometry and biophysics

soma { L = 50 // um length diam = 50 nseg = 1 insert hh // HH mechanism}

dend { . . .

from neuron import hSection = h.Section

soma = Section()dend = Section()dend.connect(soma, 1, 0)

soma.L = 50soma.diam = 50soma.nseg = 1soma.insert('hh')

dend { . . .

0 1 2 3 0.00

0.20

0.40

0.60

0.80

1.00

t

Backward Euler

y(t + dt) = y(t) + dt *f(y(t + dt))

dt = .75

y’ = f(y)

=y(t+dt) − y(t)

dtf(y(t+dt))

0 0.2 0.4 0.6 0.8 10

0.5

1

1.5

2

0 0.2 0.4 0.6 0.8 10

0.5

1

1.5

2

Cvode.atol(1e−3)

Cvode.atol(1e−1)

Parallel simulations with NEURON

1. Embarrassingly parallel problems--many runs, each of which takes at least 1 second� optimization, parameter space exploration

� communication overhead < single run time

� "bulletin-board" ("master/worker") approach

� little programming effort required

2. Multithreaded execution� shared memory architecture

(e.g. multicore PC/Mac)

� often no programming effort required

Parallel simulations with NEURON

3. Simulation of a large model that is distributed over multiple processors� spike source and target may be on different

processors

� a large cell may be distributed over multiple processors

� significant programming effort is often required to recast serial model implementation into form suitable for parallel simulation

� effort is repaid only if you have to execute many runs, each of which takes a long time (hours)

Parallel simulations with NEURON

Speedup is nearly proportional to number of processors as long as each processor has at least ~200 ODEs to integrate.

Migliore, M, Cannia, C., Lytton, W.W., Markram, H. and Hines, M.L. Parallel network simulations with NEURON. Journal of Computational Neuroscience 21:119-129, 2006.

Hines, M.L., Markram, H. and Schuermann, F. Fully implicit parallel simulation of single neurons. Journal of Computational Neuroscience 25:439-448, 2008.

Kumar, S., Heidelberger, P., Chen, D. and Hines, M. Optimization of applications with non-blocking neighborhood collectives via Multisends on the Blue Gene/P supercomputer. 24th IEEE International Parallel and Distributed Processing Symposium, in press, 2010.

Preprints available at http://www.neuron.yale.edu/neuron/nrnpubs