Cx3Dp: A Parallel Framework for Modeling the Growth and Development of Neural Tissue

Andreas Hauri (Institute of Neuroinformatics ETHZ/UZH), Rodney Douglas (Institute of Neuroinformatics ETHZ/UZH)

The development of neural tissues such as the neocortex unfolds from a limited number of stem cells by successive mitosis, migration, differentiation and adaptation of cells that finally come to take up specific functional roles. Unlike self-assembly in which the necessary components are given, and assembly depends largely on complementary structure and forces of interaction. The self-construction of development makes explicit use of structural and organizational information encoded in the genome. We have been exploring this complex process by modeling and simulation. In a previous publication, we described a software framework (Cx3D), that enables the simulation of development in physical 3D environment that respects the physical of interaction of objects as well as diffusion of morphogens. (Zubler and Douglas, 2009). However, the success of that system in simulating development soon brought us to the limits of single-threaded, single-computer applications. To open the route to simulating larger cortical areas and their inter-areal connections, we require a more general version of Cx3D that can exploit multiprocessor/multicomputer systems. To this end we have designed a parallel version of  Cx3D (Cx3Dp) that addresses these improvements and can also be run cross-platform. The system scales well with addition of processors and computers. Computers can be dynamically added to the simulation if the need for more computational power or more memory arises. The load balancing will acquire the new resources and balance the system in a way that each computer takes the same amount of time to complete one time step. The simulation can be saved after any time step, over multiple computers, and can be reloaded and continued again if necessary. The diffusion framework has also been redesigned. The diffusion is now implemented on an octree grid system that can be adapted to a desired resolution, according to the resolution necessary for the diffusion. Compared to the previous version of Cx3D. This improvement allows higher flexibility in assigning the computational resources to the diffusion. Currently we are running simulations of up to 200’000 cells and 400’000 cell compartments on 3 machines with 64 GB of memory and 24 cores each,  using only 25% of the memory available. The system scales well in speed when adding cores, and in speed and memory on adding machines.  The new Cx3Dp thus opens the way to very large scale simulation of developing systems on simple networks of consumer computers.

Preferred presentation format: Demo
Topic: Large scale modeling

Document Actions