Supercomputing Institute Research Bulletin Fall 1997

Heterogeneous Computing Applications Answer Questions in Neuroscience

A computer simulation of unprecedented scale in the field of neuroscience is the result of an international collaboration which includes Supercomputing Institute researchers. The project attempts to answer critical questions about the function of the human brain. Specifically, the researchers are developing software that will assist in the combinatorial exploration of the parameter spaces that describe single neurons. The research group includes Supercomputing Institute Fellow and Professor of Pharmacology George Wilcox, project director and visiting research scholar Rogene Eichler West, Shep Smithline and Andy Hollenbeck of Minnesota Supercomputer Center Inc., and summer interns Erik Johnson and Matt Anderson. They plan to demonstrate the simulation in late winter or early spring. The project is based on earlier research conducted by Professor Erik DeSchutter of the University of Antwerp, Belgium. DeSchutter found that under experimental observation, neurons fire spikes at different rates, depending on their environment. DeSchutter sought to determine how morphology and the distribution of protein channels in the neuron membrane account for this variability. However, technical constraints limit the ability to manipulate these variables experimentally. A simulation-based combinatorial approach, therefore, is an ideal method to answer these questions. To create a simulation of neuron firing activity, the researchers designed software that contains five interacting modules: an evolutionary algorithm (EA), a numerical simulator (GENESIS), custom graphics (NeuronViz), statistical analysis programs, and a scheduler, which the researchers are developing so that it may be also used by other Institute researchers. The simulations will be performed simultaneously on the Cray C90, Cray T3E, IBM-SUR cluster, and the SGI R10000 and Origin installations on the University of Minnesota campus. The simulation’s output is the neuron’s transmembrane voltage at thousands of points in time and at thousands of positions in the neuron’s complex dendritic tree. A playback of these points describes the neuron’s electrical behavior. Each simulation requires a minimal amount of memory (less than 20 megabytes). However, the run times are quite long—it takes more than an hour to evaluate each of the parameter sets. Fortunately, since each parameter set can be evaluated independently of the other sets, the application is strikingly parallel and requires minimal load balancing capabilities. The researchers are using a neuron model developed by DeSchutter and his colleagues, called the DeSchutter-Bower Purkinje model, that consists of a 32-dimensional parameter space. To explore this space, the group must run tens of thousands of simulations, but they are able to learn much about the sensitivity of the model to parameter selections and about the many parameter sets that yield behavior close to experiment. The combinatorial/computational approach to investigating neural function will likely mature over the next decade, yielding a cost-effective means of evaluating the effects of pharmacological agents on the behavior of neurons. Graphic representations of the voltage in each of about 1,000 compartments of a Purkinje neuron, simulated with the GENESIS neuronal simulator. Each image is a snapshot of a point in time during the simulation. The colors represent voltage along a continuum where red represents +40mV and blue represents–70mV.

In This Issue: 1997 Research Scholars LCPC Workshop T3E Upgrade > Computing Applications in Neuroscience Unraveling Protein Structures Silicon Nanocrystals Research Reports