III: Small: Collaborative Research: Supporting Efficient Discrete Box Queries for Sequence Analysis on Large Scale Genome Databases

该合作研究项目由密歇根州立大学（MSU）和密歇根大学迪尔伯恩分校（UM-D）的研究人员共同开展，研究了用于生物信息学序列分析的大规模k-mer数据集（即从基因组序列中获得的重叠k-长度子序列）的存储和搜索/查询问题和技术。高效的k-mer索引、存储和检索对于序列分析任务（如纠错）至关重要，因为测序数据集的大小会急剧增加。大多数现有的存储和搜索k-mer的方法都针对精确查询或范围查询进行了优化。然而，这种依赖限制了可以有效完成的序列分析类型。此外，大多数现有的k-mers存储方法都不支持多字长的k-mers高效存储。对于许多序列分析问题，包括纠错、变异检测和组装，使用多个单词长度进行搜索可以提高灵敏度和特异性。在这个项目中，研究了在大规模k-mer数据集上有效支持所谓的（离散）盒查询和其他相关查询（例如混合查询）的各种技术，用于序列分析。研究了在解决诸如纠错等序列分析问题时优化盒查询的方法。探讨了k-mer数据集上支持多字长搜索的存储结构和框查询的采用。本研究结果将推动基因组序列数据库存储、索引和检索技术的发展。它们有望通过为序列分析提供新的高效磁盘解决方案，显著影响生物信息学的当前实践。它们还将影响许多其他流行的应用领域，包括生物识别、图像处理、社交网络和电子商务，在这些领域中，处理无序离散多维数据至关重要。该合作研究项目由密歇根州立大学（MSU）和密歇根大学迪尔伯恩分校（UM-D）的研究人员共同开展，研究了生物信息学中用于序列分析的大规模k-mer数据集的存储和搜索/查询问题和技术。高效的k-mer索引、存储和检索对于序列分析任务（如纠错）至关重要，因为测序数据集的大小会急剧增加。大多数现有的存储和搜索k-mer的方法都针对精确查询或范围查询进行了优化。然而，这种依赖限制了可以有效完成的序列分析类型。此外，大多数现有的k-mers存储方法都不支持多字长的k-mers高效存储。对于许多序列分析问题，使用多个字长进行搜索可以提高灵敏度和特异性。在这个项目中，研究了在大规模k-mer数据集上有效支持所谓的（离散）盒查询和其他相关查询（例如混合查询）的各种技术，用于序列分析。特别地，开发了一种新的索引树，命名为BoND-tree，专门为以k-mer数据集为特征的非有序离散数据空间设计。利用空间的独特性质，开发了新的索引树节点分割启发式算法，并进行了理论分析，证明了所提启发式算法的最优性。除了基于数据分区的BoND-tree外，还开发了基于空间分区的框查询索引方案。为了支持更灵活的查询类型（即混合盒查询和范围查询），研究了结合盒查询索引和范围查询索引优点的混合索引方案。为了促进大规模k-mer数据集的有效索引构建，还开发了针对所提出的索引树的批量加载技术。此外，本文还研究了优化盒查询在解决诸如纠错等序列分析问题中的方法。本文还探讨了k-mer数据集上支持多字长搜索的存储结构和框查询的采用。该项目的研究将导致发现生物信息学中序列数据的数据空间的基本属性，开发一些新的存储，索引和检索技术，利用这种数据空间的属性，并应用所提出的技术来解决序列分析中的重要问题。这些结果将推动基因组序列数据库存储、索引和检索技术的发展。它们有望通过为序列分析提供新的高效磁盘解决方案，显著影响生物信息学的当前实践。它们还将影响许多其他流行的应用领域，包括生物识别、图像处理、社交网络和电子商务，在这些领域中，处理无序离散多维数据至关重要。

This collaborative research project, conducted jointly by the investigators from the Michigan State University (MSU) and the University of Michigan at Dearborn (UM-D), investigates the issues and techniques for storing and searching/querying large scale k-mer data sets (i.e., overlapping k-length subsequences obtained from genome sequences) for sequence analysis in bioinformatics. Efficient k-mer indexing, storage and retrieval are vital to sequence analysis tasks like error correction as sequencing data set sizes increase vastly. Most existing methods for storing and searching k-mers are optimized for exact or range queries. However, this reliance limits the types of sequence analysis that can be done efficiently. Moreover, most existing methods for storing k-mers do not support efficient storage of k-mers at multiple word lengths. For many sequence analysis problems, including error correction, variant detection, and assembly, searches with multiple word lengths enable better sensitivity and specificity. In this project, various techniques for efficiently supporting so-called (discrete) box queries and other related queries (e.g., hybrid queries) on large scale k-mer data sets for sequence analysis are investigated. The approaches to optimizing box queries in solving sequence analysis problems like the error correction are examined. The storage structure and adoption of box queries for supporting searches with multiple word lengths on k-mer data sets are explored. The results from this research will advance the state of knowledge for storage, indexing and retrieval techniques for genome sequence databases. They are expected to significantly impact current practice in bioinformatics by making available new efficient on-disk solutions for sequence analysis. They will also impact a number of other popular application areas including biometrics, image processing, social network, and E-commerce, where processing non-ordered discrete multidimentional data is crucial. This collaborative research project, conducted jointly by the investigators from the Michigan State University (MSU) and the University of Michigan at Dearborn (UM-D), investigates the issues and techniques for storing and searching/querying large scale k-mer data sets for sequence analysis in bioinformatics. Efficient k-mer indexing, storage and retrieval are vital to sequence analysis tasks like error correction as sequencing data set sizes increase vastly. Most existing methods for storing and searching k-mers are optimized for exact or range queries. However, this reliance limits the types of sequence analysis that can be done efficiently. Moreover, most existing methods for storing k-mers do not support efficient storage of k-mers at multiple word lengths. For many sequence analysis problems, searches with multiple word lengths enable better sensitivity and specificity. In this project, various techniques for efficiently supporting so-called (discrete) box queries and other related queries (e.g., hybrid queries) on large scale k-mer data sets for sequence analysis are investigated. In particular, a new index tree, named the BoND-tree, specially designed for a non-ordered discrete data space characterized by k-mer data sets is developed. The unique properties of the space are exploited to develop new node splitting heuristics for the index tree, and theoretical analysis is performed to show the optimality of the proposed heuristics. Besides the BoND-tree, which is based on data partitioning, space-partitioning based index schemes for box quieres in such a space are also developed. To support a more flexible type of query (i.e., hybrid box and range queries), hybrid index schemes integrating strengths of both box query indexes and range query indexes are studied. To facilitate an efficient index construction for large scale k-mer data sets, bulk loading techniques are also developed for the proposed index trees. In addition, the approaches to optimizing box queries in solving sequence analysis problems like the error correction are examined. The storage structure and adoption of box queries for supporting searches with multiple word lengths on k-mer data sets are also explored. The research in the project will result in the discovery of fundamental properties of the data space for sequence data in bioinformatics, the development of a number of novel storage, indexing and retrieval techniques exploiting the properties of such a data space, and the applications of the proposed techniques for solving important problems in sequence analysis. These results will advance the state of knowledge for storage, indexing and retrieval techniques for genome sequence databases. They are expected to significantly impact current practice in bioinformatics by making available new efficient on-disk solutions for sequence analysis. They will also impact a number of other popular application areas including biometrics, image processing, social network, and E-commerce, where processing non-ordered discrete multidimentional data is crucial.