BrainGrid

Student research opportunities available for more information, see our student projects page.

Some of the central questions in neuroscience must include how the activities and connectivities of individual cells in cortex contribute to its development and operation as a computational mechanism. A powerful experimental approach for investigating these questions is the use of cultured dissociated cortical cells grown into networks on a multi-electrode array. Such preparations allow investigation of network development, activity, plasticity, responses to stimuli, the effects of pharmacological agents, etc.

A common behavioral feature of such preparations is the occurrence of whole-culture pathological (in the sense that it does not occur in vivo) bursting that oftentimes interferes with the experimental goals. This bursting is interesting from both a theoretical point of view, as well as a clinical one (as it has been proposed as a model for epilepsy). Understanding the mechanisms that underlie bursting could allow creation of more useful cell cultures and possibly have medical applications.

In this project, we are performing a computational study of the interplay of individual neuron activity, cell culture development, and the network behavior. This is very large computing task, as we need to simulate each neuron's activity at sub-millisecond accuracy while performing simulations of weeks of cortical development. Moreover, we must do this for networks of thousands to tens of thousands of cells.

Our approach to answering this computing need is to use general-purpose graphics processing unit (GPU) hardware to accelerate the "micro-level" simulation steps (the sub-millisecond integration at the individual neurons), leaving to the CPU the tasks of growth and development and inter-neuron communications. Along the way, we hope to develop a library that can provide researchers performing neural simulations with a head start on using GPU hardware.

The name "BrainGrid" comes from a previous investigation of the use of agent-based middleware (AgentTeamwork) for this application. While the current approach isn't really grid computing anymore, we liked the name.

While we work on the large-scale simulation, we have been developing smaller-scale, uniprocessor versions as proof of concept and for validation purposes for the parallel code. Some early results were presented at the 2007 International Workshop on Neuronal Coding, in Montevideo, Uruguay.