全局基因调控网络构建与分析并行算法分析.pdfVIP

PARALLEL ALGORITHMS FOR CONSTRUCTING AND ANALYZING THE GENOME-WIDE GENE REGULATORY NETWORKS ABSTRACT As the researchers’ focus has shifted to function. This make the construction and analysis of genome-wide gene regulatory networks become a significant topic of computational biology. While the construction and analysis of genome-wide gene regulatory networks needs too much time complexity, how to use the CPU-GPU collaborative computing to solve the problem has become a new research topic. Based on the features of CPU/GPU mixed heterogeneous computing platforms, this paper designed a parallel algorithm which has high memory access efficiency, high speed ratio and good scalability for constructing genome-wide gene regulatory networks. First, the algorithm divide gene expression profiles into basic blocks and calculate by dividing the diagonal rounds to reduce the data memory access conflicts. Second, allocate computing tasks by the computing power of CPU and GPU to ensure load balancing. For the CPU side, the algorithm further divided the basic block gene expression profiling reasonable into sub-blocks by multi-level cache size. The reasonable dispatched sub-block to the multi-level cache can ensure a high cache hit rate III and reduce the number of data memory access exchange. For the GPU side, we designed a double diagonal parallel multi-GPU to increase hundreds of core utilization while reducing memory access conflicts based on structural characteristics, improving efficiency of the algorithm by using texture memory cache to achieve efficient memory access ef