BIGDATA: Low-Memory Streaming Prefilters for Biological Sequencing Data

描述（由申请人提供）：我们将很快能够对整个细菌群落以及每个肿瘤细胞的DNA和RNA进行详尽的测序。这两种非常不同的测序应用都需要快速有效地处理大量有噪声的序列数据（几十到100兆兆的数据库），以从噪声中分离信号并产生生物学洞察力。然而，目前用于从这些数据中提取信息的生物信息学方法无法轻松处理所获得的大量数据。处理该序列数据的主要挑战是双重的：相对较高的错误率（每个碱基0.1- 1%），以及我们现在可以使用测序仪（如lllumina HiSeq）轻松获取的数据量。多年来，测序能力每6个月翻一番，比计算能力快得多。由于几乎所有现有的生物信息学分析方法都需要多次遍历原始数据，并且许多分析算法尚未并行化，因此生物信息学分析能力继续远远落后于数据生成能力。此外，许多现有的软件包不能轻易地重组以利用许多核心或GPU算法，因此不会利用我们提出的在散弹法测序数据的损失压缩和错误连接中开发和实现新的流媒体方法的计算能力和网络基础设施的预期进步。我们的算法是少通道（$<$ 2），不需要特定于样本的信息，并且可以在固定或低内存中实现；此外，它们支持并行化，可以在许多核心环境中高效运行。当作为现有分析包的预过滤器实现时，我们的方法将消除或纠正数据集中的大部分错误，大大减少了使用现有包进行下游分析的计算空间和时间要求。此外，我们将通过扩展mRNAseq和宏基因组数据集的纠错方法来提供新的能力。智力优势：我们将开发一系列的算法空间和时间效率压缩和错误纠正短读DNA和RNA序列数据。这些策略将大大提高许多下游分析应用的可扩展性，从宏基因组的群落分析到人类的重测序分析。我们将提供描述空间和时间效率和灵敏度之间权衡的分析，并提供经过测试的、文档化的参考实现，社区可以使用这些方法进行实际评估并将其纳入分析工具中。我们的方法将通过为两种最常见的短读序列分析（映射和组装）引入高效和有效的流方法，显著影响短读序列分析。

DESCRIPTION (provided by applicant): We will soon be able to exhaustively sequence the DNA and RNA of entire communities of bacteria, as well as every individual cell of a tumor. Both of these very different applications of sequencing share in the need to rapidly and efficiently sor through large amounts of noisy sequence data (dozens to 100s of terabases) to separate signal from noise and produce biological insight. However, current bioinformatics approaches for extracting information from this data cannot easily handle the vast amounts of data being acquired. The primary challenges in processing this sequence data are twofold: the relatively high error rate of 0.1-1\%, per base, and the volume of data we can now easily acquire with sequencers such as lllumina HiSeq. For years, sequencing capacity has been doubling every 6 months -significantly faster than compute capacity. Since almost all extant bioinformatics analysis approaches require multiple passes across the primary data, and many analysis algorithms have not been parallelized, bioinformatics analysis capacity continues to lag ever further behind data generation capacity. In addition, many of the existing software packages cannot easily be retooled to take advantage of many core or GPU algorithms, and hence will not take advantage of expected advances in compute capacity and cyber infrastructure we propose to develop and implement novel streaming approaches for loss compression and error connection in shotgun sequencing data. Our algorithms are few-pass ($<$ 2), require no sample-specific information, and can be implemented in fixed or low memory; moreover, they are amenable to parallelization and can run efficiently in many core environments. When implemented as a prefilter to existing analysis packages, our approaches will eliminate or correct the majority of errors in data sets, dramatically reducing the computational space and time requirements for downstream analysis using existing packages. Moreover, we will provide novel capability by extending error correction approaches to mRNAseq and metagenomic data sets. Intellectual Merit: We will develop a range of algorithms for space- and time-efficient compression and error correction of short-read DNA and RNA sequence data. These strategies will substantially increase the scalability of many downstream analysis applications, ranging from community analysis of metagenomes to resequencing analysis of humans. We will provide analyses describing the tradeoffs between space and time efficiency and sensitivity, and deliver tested, documented reference implementations of our approaches that can be used by the community for practical evaluation and incorporation into analysis tools. Our approaches will significantly impact short-read sequence analysis by introducing efficient and effective streaming approaches to the two most common types of short-read analysis, mapping and assembly.