(go to slideshow)

Realistic Modeling of Large Networks of Neurons

Piotr J. Franaszczuk, Pawel Kudela, Gregory K. Bergey
 
Department of Neurology, Johns Hopkins University School of Medicine and Epilepsy Center,
Johns Hopkins Hospital , Baltimore, Maryland U.S.A.


The realistic computational models of neural networks based on knowledge of anatomy and physiology are
becoming increasingly utilized in neuroscience. Computer resources impose limitations on the level of detail
that can be included in the models.

We present here a neural simulator designed for simulations of a large network of realistic neurons.
Each neuron is modeled as a single compartment H-H type model with additional synaptic currents. Synaptic
currents are modeled as dual-exponentials with specific weight and delay parameters. Modification of the
synaptic current parameters allows for the simulation of effects of different properties as well as
different locations of synapses on the dendritic tree. All neurons are divided into classes sharing the same
physiological properties. This allows for optimization of memory for each neuron and synapse. The current
version of the simulator allowing for simulations of up to 65536 neurons uses approximately 12k bytes per
class of neurons, 128 bytes per neuron and 2 to 6 bytes per synapse.

The topology of network connections for large networks is generated using a pseudo-random generator
based on the average number of synapses per neuron and the range of potential connections.  Output from the
simulator consists of traces of membrane potentials, ion concentrations, current conductivities from
selected neurons and histograms of action potentials from all neurons. The latter can be displayed as an
animated movie.

The simulator was successfully used to model traveling waves of epileptiform activity in neocortical
neurons.

(Supported by NIH grant NS 38958)



(go to slideshow)

(go back)