Authors: Sohan Lal, Kolin Paul, James Gomes
Abstract—The complexity of metabolic networks and their regulation is a very crucial research area for system biology. The advances in biochemistry and molecular biology have produced an enormous amount of data which are used in mathematical models to simulate complex pathways. The models are computationally very expensive — however, most of these exhibit a large degree of parallelism. Recently, graphics processing units (GPUs) have become a handy tool to build accelerators for data parallel applications. In this paper, we propose a methodology for simulating the metabolic networks on general purpose graphics processing unit (GPGPU) starting from a SIMULINK description
of the network. The network is systematically ported to run on a GPU using the NVIDIA’s CUDA (Compute Unified Device
Architecture) environment. The high degree of parallelism allows a significant speedup of the CUDA implementation compared to the generic SIMULINK implementation. We got a speedup of more than 6000X of CUDA implementation over the SIMULINK implementation when a metabolic network of 1024 nodes is simulated for one minute with a step size of 0.01 minute. We have also studied the relationship between the step size, which is used to solve the delay differential
equations representing the dynamics of network nodes and the delta simulation time, which is the time when the processing nodes exchange information according to network. This results in an interesting relationship between accuracy and speed of simulation.