基于GPU加速的基因组数据压缩框架.pdfVIP

基于 GPU 加速的基因组数据压缩框架 1,2* 1,2* 1,2+ 1+ 2 郭贵鑫 , 邱爽 , 王丙强 , 方林 , Simon See 1 深圳华大基因研究院,深圳 518083 2 华大基因-英伟达联合创新实验室,深圳 5 18083 * 同等贡献作者 GPU-Accelerated Compression Framework for Genomic Data 1,2* 1,2* 1,2+ 1+ 2 GUO GuiXin , QIU Shuang , WANG BingQiang , FANG Lin , SEE Simon 1BGI Shenzhen, Shenzhen 518083 2BGI-NVIDIA Joint Innovation Lab, Shenzhen 518083 * These authors contribute equally to this work + Corresponding author: Phn +86-755-2235-5697, E-mail: wangbingqiang@, Received 2000-00-00; Accepted 2000-00-00 Abstract : Genomic data is incoming at unprecedented rate, with most of the genomics data files stored as table-like plain text. To facilitate data storage in terms of both volume and balanced infrastructure, we develop a GPU-accelerated compression framework. The work combines generic compression algorithms and GPU acceleration to achieve compression / decompression rate around 100MB/s without compromising compression ratio, through selection of appropriate compression scheme for a given column of data. We further present a novel first-order Markov model based transformation. Experiments show that in some contexts this method is an excellent replacement of Burrows-Wheeler and Move-To-Front transformation. Key words: GPU; data compression; parallel algorithm; Markov model 摘 要: 基因组数据在以空前的速度增长，其中，多数基因组数据文件格式以类似于表格的平文形式存储。为了 合理利用资源，我们开发了一个基于GPU 加速的基因组数据压缩框架。该框架包含了多种压缩算法在GPU 上的 高效并行化实现，通过为列数据选取合适方法进行压缩，在不损失压缩率的前提下，压缩速度和解压速度同时达 到约100MB/s。同时本文设计了一种适合GPU 加速的基于Markov 模型及字符变换的压缩方法，测试结果表明该 方法在某些文本环境下可以很好地替代基于Burrows-Wheeler 和Move-To-Front