Accelerating the smoldyn spatial stochastic biochemical reaction network simulator using GPUs

Smoldyn is a spatio-temporal biochemical reaction network simulator. It belongs to a class of methods called particle-based methods and is capable of handling effects such as molecular crowding. Individual molecules are modelled as point objects that can diffuse and react in a control volume. Since each molecule has to be simulated individually, the computational complexity of the simulator is quite high. Efficiently executing high fidelity models (> 106 molecules) is not feasible with traditional serial computing on central processing units (CPUs). In this paper we present novel data-parallel algorithms designed to execute on graphics processing units (GPUs) to handle the computational complexity. Our preliminary implementation can handle diffusion, zero-order, uni-molecular, and bi-molecular reactions. Our preliminary results show performance gain of over 200x over the original implementation without loss of accuracy.