GENESIS is a package designed to allow the construction of a wide variety of neural simulations. It was originally designed by Matt Wilson at Caltech to assist in his doctoral modelling work on the Piriform Cortex [Wilson:89b]. One of the design objectives was to allow the easy construction and alteration of a wide variety of neural models from detailed single cells all the way up to complex multilayered neural network structures. In order to make the simulator as flexible as possible, it was decided to adopt an object-oriented approach to the underlying simulator and to allow the user to include precompiled libraries of elements appropriate to their particular scale of modelling (e.g., detailed single cells or network-scale models). The structure of individual models was described via neural description script language files, which were interpreted as execution of the model proceeded. This combination of interpreted script files and precompiled element libraries has proved to be a very powerful approach to the problems of neural modelling at a variety of levels of detail from detailed single cell models all the way up to large network-scale simulations composed of thousands of neural elements.
GENESIS is an object-oriented neural simulator. All communication between the elements composing the simulation is via well-defined messages. As such, it was expected to fit well into the distributed computing environment of modern parallel computers.
It is designed in an object-oriented manner where each GENESIS element has private internal data that other elements cannot access directly. They can only access this information via predefined messages that request the internal state information from an element.
The GENESIS neural simulation system has now been running successfully on two of the Intel parallel computers at Caltech since 1991 and has already produced biologically interesting and previously unobtainable results. Much of the use of the simulator to date has been in the construction of a highly detailed model of the Cerebellar Purkinje Cell (work produced by Dr. Erik de Schutter at Caltech using the parallel GENESIS system provided by ourselves) [Schutter:91a;93a]. This is thought to be one of the most detailed and biologically realistic single-cell models developed to date. By utilizing the special capabilities of the Parallel GENESIS system, it has been possible to carry out simulated experiments which are presently very difficult to carry out experimentally. The initial results have been very promising and have shown several previously unsuspected properties of the Purkinje Cell, which arise as a result of the anatomical and physiological properties of the cell's dendritic tree. The ability to run up to 512 different Purkinje Cell models simultaneously has allowed the construction of statistically significant profiles of Purkinje Cell response patterns. This research has previously been impossible to conduct for detailed cell models because of the excessive computational power required. Until now, the only statistical behavior that has been described is for population dynamics of very simplified neural elements.
Currently, these machines are being used in two distinct ways (the task farming approach and the distributed model via the postmaster element) [Speight:92a;92b].