Chapter 1. Introduction

"The purpose of computing is insight, not numbers" -- R. Hamming

While some analytic mathematic expressions for nerve (and muscle) membrane properties preceded the monumental work of Hodgkin & Huxley, it was their equations for the ionic channels in the membrane of the giant axon of the squid which set computational neuroscience in motion. Since then not only have variations of sodium and potassium channels been described but many other types of ion channels have been found. Today molecular neurobiologist are making a great variety of hybrid and "designer" channels, leading to great strides in understanding the molecular features underlying their properties.

In this period our knowledge of the detailed morphology of many nerve cells has been vastly increased by an array of technological improvements. Video techniques have revived light microscopy by improving contrast and "resolution". This, along with computer databases of geometric details for components of neurons, allows a full 3-dimension reconstruction, including the location of dendritic spines and synaptic inputs. Electron microscopy can provide finer details where necessary.

More recently, the speed and memory capacity and the availability of desktop computers has been exploding. The technological advances in the manufacture of their components has helped reduce their price and the resulting increase in purchases has driven automation of high volume chips; the resulting positive feedback allows further cost reduction.

Concurrent with these developments in hardware, computer software has been advancing rapidly because the volume of sales has allowed the investment in powerful programming tools. This, in turn, with the large volume of sales allows the price of software to be reduced. The net result is a very rapid growth in the power and speed of computer programs. For the presently very limited field of computation neuroscience the emergence of the software development tools has allowed the construction of several powerful simulation tools. A group of neural simulators were demonstrated in an exhibit booth at the 1993 meeting of the Society of Neuroscience:

DSTOOL by John Guckenheimer, Cornell Univ., dynamical systems on Unix machines
GENESIS by Jim Bower, Cal. Tech., general purpose simulator for neural systems on Unix machines
NBC by Jean-Francois Vibert, Fac. de Med. St-Antoine, Paris, Network simulation and analysis on Unix and VMS machines
NEMOSYS by John Tromp, Univ. Cal., Berkeley, complex single neurons on Unix machines
NEUROGRAPH by Peter Wilke, Univ. Erlangen, Germany, Simulation of artificial neural networks on Unix, DOS, VMS machines
NEURON by Michael Hines, Duke Univ., Simulations of biologically realistic single neurons and small networks on PCs and Unix machines
NEURONC by Rob Smith, Univ. Penn., compartmental simulations of large neural circuits on Unix machines
NODUS by Eric De Schutter, Univ. Antwerp, Belgium, simulation of small networks of neurons on Macintosh machines
NSL by Alfredo Weitzenfeld, Univ. Sou. Cal., simulation of large networks on Unix machines
SNNAP by John Byrne, Univ. Texas, Houston, Simulator for neural networks on Unix machines
SWIM by Orjan Ekeberg, Royal Inst. Tech., Stockholm, simulation of network of few compartment model neurons on Unix machines

Now, at the time of this writing, the early 1990s, we see that the convergence of software tools, computer power, morphology, and ionic channel data can allow us to carry out meaningful simulations of full neurons. Actually the present computer and software power are sufficient to carry out simulations of nerve networks in great detail. The morphological detail of several neuron types is available. The major problem in making definitive simulations is the lack the necessary detailed information on the locations and densities of the various types of channels and other membrane mechanisms. Thus the only way to proceed is to employ informed guesses and to attempt to fit electrical records from different kinds of experiments (e.g. voltage clamp, current clamp of the soma, orthodromic stimuli, antidromic invasion).

A graphical representation of my perception of the history of the increase in knowledge of ionic channels and neuron morphology overlaid with that of computer hardware and software.

It seems to me that we are on the verge of new era of insight into the functioning of nerve cells and networks thereof which will be gained by using all of these tools in an environment of tight coupling between experimentation and modeling. The purpose of the development of the simulation program NEURON and the writing of this manual is to make a contribution to this effort.

Waves of the Future:

Another prospect on the horizon is the use of advances in computer technology and networks to work together as a community of scholars, pooling our databases, tools, and results. This cooperation would enormously enhance one's ability to create much better realistic simulations in a far more efficient manner. The present informal communication between Mike Hines, the designer of NEURON, and its users needs to be replaced with a more formal mode allowing more direct communication within the user community and dramatically reduce repetition by users without every inquiry being funneled through Hines. We hope to accomplish this by maintaining a registry and database on a machine dedicated to NEURON which would contain the names of users, their addresses, problems, publications & abstracts, simulation figures, and templates of their programs.

Libraries of Contributions

Several users have offered contributions (results from their work) to enhance the use of NEURON. For example, one user has offered his library of some 50 ionic channel types, complete with references, equations, etc. This example shows how a public domain enterprise can enhance and speed the work of the whole community by avoiding duplication of effort. We envision a continuing flow of contributions from an enlarging user group. Hines will review and test each contribution for quality, accuracy, etc. before inserting it into the libraries of channels, mechanisms, additional features.


A variety of types of data are necessary for realistic simulations, for example:
  • Morphology of neurons reconstucted from microscopic analysis. NEURON already has a filter to input data from Eutectic and other measurement systems.
  • Synapses: types, locations, strengths, and sources as well as temporal patterns of inputs
  • Channels: voltage-sensitivity, calcium, second-messangeer, and pharmological sensitivities. Density profiles in nerve cells.
  • Receptors: types, density profiles, modulation by second-messanger
  • Intracellular Milieu: Ca++ stores and binding proteins, kinases, phosphotases
  • Ca++ imaging: [Ca] as a function of stimulus, time, and location
  • Electrical records for current clamps: voltage responses to step currents, epsps, orthodromic and antidromic stimuli
  • Electrical records for voltage clamps: current responses to step voltages, epsps, orthodromic and antidromic stimuli
  • Single channel data: conductances, kinetics of openings/closings, modulations by kinases & phospohtases. Molecular component manipulations, mutations


Publication of simulations in conventional journals has long been unsatisfactory because the space limitations preclude the publication of the methods, parameters and other details necessary for a reader to reproduce the author's results. With the present possibilities for electronic storage of a program, it would be sufficient to have a brief version published in a conventional journal with the essential results and conclusions combined with full simulation details (such as all parameter values, sensitivity studies, output curves) stored electronically for access via the internet. (Such an approach is now offered by the Biophysical Journal via a gopher server). The quality of these electronic documents could be judged by a small panel of experts - as with conventional journals - or by a rating derived from the evaluation of a large number of users. Those programs/publications which receive a high rating can be should be stored in a repository library for others to use as templates and again reduce unnecessary repetition. Thus we could provide excellent examples of simulations of a variety of neuron types and networks. This process, coupled with e-mail or bulletin board announcements would represent the ultimate speed of "rapid publication"!

Distributed and Parallel Computer Processing

As our simulation problems become larger and more complex with large numbers of very detailed nerve cells in a neural network, high speed information networks, linking an array of computers will offer a way to distribute the processing over many CPUs and thus carry out simulations in reasonable times. Similarly, as parallel computing algorithms mature, the computational load can be shared betwen many processors in a single machine.

Last update: 9/12/95