Genome Informatics For Biobank-scale Data

生物库规模的遗传数据提供了丰富的信息，这些信息在传统的生物库中并不明显。 更小的队列。我们将开发和评估有效和准确的算法和工具， 分析这些数据以揭示这些信息。特别是，我们将开发三种方法 主要任务：单倍型定相细化、基因型插补和相关性推断。 尽管在传统的较小数据集中，这些任务有成熟的方法， 仍然缺乏可扩展的有效和准确的方法和工具来处理基因组大数据， 这个规模。我们的主要观察结果是，生物库规模的遗传数据提供了密集的联系， 在各个数据点之间。与传统方法不同的是， 马尔可夫模型（Hyndrome），我们使用个体特异性队列对每个个体进行建模，即，所有 与这个人有联系的其他人我们利用高效的位置 Burrows-Wheeler变换（PBWT），用于建模单体型的基础数据结构 匹配.我们是第一个开发基于PBWT的方法来识别IBD片段的人。 生物库规模的队列研究我们还丰富了传统的PBWT数据结构， 算法，以有效的单体型搜索和允许动态更新。在本申请中，我们 利用我们的算法开发专业知识，开发基于IBD的算法， 非常大的面板的单倍型定相。我们还将开发基于IBD的算法， 使用大量参考文献提高基因型估算的效率和成本效益 面板此外，我们还将开发RaPID-Affin算法，以有效和准确地推断 遗传相关性最后，我们将对这些方法进行基准测试，并为 社区该项目将通过开发高效的基因工程技术， 用于非常大的基因分型队列的信息学工具。

Genetic data of biobank scales offer a wealth of information that is not obvious in traditional smaller cohorts. We will develop and evaluate efficient and accurate algorithms and tools for the analysis of such data to reveal such information. In particular, we will develop methods for three main tasks: haplotype phasing refinement, genotype imputation, and relatedness inference. Although mature methods are available for these tasks in traditional smaller data sets, there is still a lack of scalable efficient and accurate methods and tools for handling genomic big data of that scale. Our main observation is that biobank-scale genetic data offer dense connections between individual data points. Unlike traditional methods based on the Li and Stephens hidden Markov models (HMMs), we models each individual using the individual-specific cohort, i.e., all the other individuals that are connected to the individual. We leverage the efficient positional Burrows-Wheeler transformation (PBWT), a foundational data structure for modeling haplotype matching. We were the first to develop a PBWT-based method for identifying IBD segments in biobank-scale cohorts, RaPID. We are also enriched the traditional PBWT data structure and algorithms to efficient haplotype search and allowing dynamic updates. In this application, we leverage our algorithm development expertise and develop an IBD-based algorithm for refining haplotype phasing of very large panels. We will also develop IBD-based algorithms for improving efficiency and cost-effectiveness of genotype imputation using a very large reference panel. In addition, we will develop RaPID-Affin algorithms for efficient and accurate inference of genetic relatedness. Finally, we will benchmark the methods and develop free software for the community. This project will empower modern genetic research by developing efficient informatics tools for very large genotyped cohorts.