The Research Of Docking's Accelerating Technology Based On GPU

Owing to the revolution of GPU architecture and the maturity of developing environment, GPU is more widely used in the field of scientific computing. GPU-based acceleration has proven to be an effective way by which a large amount of biological information can be processed efficiently.As one of the research contents in bioinformatics, molecular docking plays an important role in drug design. The technology of molecular docking can help to understand the pathogenesis, as well as search out the method for preventing the disease and developing specific medicine. It can be expected that the drug design's period will be shortened and the level of drug design and the ability of handling the serious and paroxysmal illness will be improved via the GPU computing under the hybrid CPU-GPU architecture for accelerating the molecular docking. This thesis analyses the open source docking software AutoDock, and implements a finely granular GPU-based parallel genetic algorithm in order to accelerate molecular docking.The innovations in this thesis include:(1) Based on the analysis of computation mechanism of genetic algorithm used by AutoDock, a parallel computing method of genetic algorithm based on CPU-GPU heterogeneous architecture for molecular docking is proposed. The method takes full advantage of GPU's computing ability by CPU and GPU's cooperation. The experimental results show that the method can accelerate the processing of docking effectively.(2) A parallel computing method based on CPU-GPU heterogeneous architecture is designed and implemented. It combines genetic algorithm and simulated annealing algorithm, accelerating the local search processing by multiprocessing-units of GPU. Comparing with genetic algorithm, the mixed algorithm can inhibit the phenomenon of premature efficiently.(3) A parallel method of generating random number is developed. The task of generating random number is divided into two parts, and each part is implemented by CPU or GPU separately. A large amount of random number is generated and transferred termly by CPU. GPU generates residuary random number. The method makes good use of CPU's computing ability. Because the random number generation and GPU computing are parallel by assigning the task between CPU and GPU effectively.