Research And Applications For CPU-GPU Collaborative Computing In Biological Data Analysis

The analysis of biological data is an extremely important process in the field of bioinformatics. Due to the explosive growth in biological data, this analysis is becoming even more time-consuming. This has seriously hampered the verification of the relevant theoretical assumptions. With CPU-GPU collaborative computing, we can build a high-performance parallel computing platform to speed up the process of data analysis with small cost.In this thesis, a CPU-GPU collaborative computing platform is formulated to deal with the problem of low hardware resource utilization and high time consuming in analysis of large biological data. After weighing technology performance and cost, we select one multi-core CPU and one GPGPU to build the platform. We implement multi-core programming with OpenMP and GPU programming with CUDA in this platform. At the same time we propose several strategies for data pre-processing and task allocation based hardware features so as to minimize time used in biological data analysis and support as well to promote the bioinformatics research.This thesis focuses on two specific biological data analysis issues:one is simulation for tagSNPs selection, the other one is DTI computing&prediction. We analysis the exsiting algorithms for tagSNPs selection and propose an improved algorithm called HTag. Then we implement HTag with parallel method in this platform, not only to decrease the number of tagSNPs, but also to reduce the running time. As DTI computing&prediction with NetCBP algorithm is highly time-consuming. We redefine and reassemble the process of the NetCBP algorithm to make it amenable to parallel computing. With parallel method, we reduce the running time when computing DTI for one single drug. In many drugs computing, we also reduce the running time with parallel task scheduling.Practical applications show that these approaches with CPU-GPU collaborative computing do shorten the processing time significantly compared with the traditional approaches. Furthermore, these approaches for data analysis and processing have potential value in other areas.