Collaborative Research: Statistical Methods and Algorithms for Genomic Data

在本研究项目中，研究人员构建了用于SNP分析的统计方法和算法，旨在增强潜在模型的生物学真实性。这个项目有三个主要目的。第一部分扩展了使用极大似然和模态推理从序列数据构建层次树的新方法的发展。树的构建是基于将这两种推理方法中的任何一种应用于祖先混合模型，该模型的参数描述了过去每个固定时间点T的种群结构。如果在一个时间点T的精细网格上估计这个结构，随着时间的推移，估计之间的关系可以图形地描述为一个层次树。第二个项目的目标是增强祖先混合模型的生物真实性，包括(a)多状态特征，(b)序列进化的先进模型，以及(c)重组。该扩展是基于使用由连续时间马尔可夫链构造的扩散核。还提出了经验贝叶斯方法，以提高整体估计精度。第三个目标是开发一种新的方法，在不知道亲本信息的情况下，从基因型数据重建单倍型序列。该方法和算法是基于祖先混合模型和多矩方法，简化了计算。此外，研究人员建议将该方法扩展到长序列，通过沿着较长的基因型序列滑动窗口，然后利用重叠估计的信息构建较长的单倍型估计。目前发布的国家生物技术信息中心（NCBI）数据库dbSNP包含超过1150万条人类单核苷酸多态性（SNP）记录，在过去4年中增加了10倍。对这些数据的分析已经成为生物信息学和计算生物学研究的一个焦点，因为它是携带信息的snp来区分物种内的个体。在这些数据中编码的是关于个体特征与其遗传密码之间关系的重要信息。研究人员正在为这些数据开发方法和模型，这些方法和模型与目前在使用SNP数据的两个领域使用的标准技术有着根本的不同，聚结和系统发育推断，其中一个是重建类似于基于当前祖先的家谱的遗传关系。该项目的更广泛影响包括开发具有广泛应用潜力的方法和模型，增加生物和数学创新的融合，以及为不同学生提供广泛的跨学科培训机会。

In this research project the investigators construct statistical methods and algorithms for SNP analysis that are designed to enhance the biological realism of the underlying models. This project has three primary aims. The first extends development of a new method for constructing hierarchical trees from sequence data using maximum likelihood and modal inference. The tree construction is based upon application of either one of these two inference methods to an ancestral mixture model, a model whose parameters describe the population structure at each fixed time point T in the past. If one estimates this structure over a fine grid of time points T, the relationship between the estimates over time can be graphically described as a hierarchical tree. The second project aim is to enhance the biological realism of the ancestral mixture model to include (a) multi-state characters, (b) advanced models of sequence evolution, and (c) recombination. The extensions are based on using diffusion kernels constructed from continuous time Markov Chains. Empirical Bayes methods are also proposed to be employed to improve overall estimation precision. The third aim is development of a new method for reconstructing haplotype sequences from genotype data without knowing the parental information. The methods and algorithms are based on the ancestral mixture models together with a multi-moment approach that simplifies computation. In addition, the investigators propose to extend the method to long sequences by sliding a window along longer genotype sequences, then using the information from the overlapping estimates to construct longer haplotype estimates.The current release of the National Center for Biotechnology Information's(NCBI) database dbSNP contains over 11.5 million human single nucleotide polymorphism (SNP) records, representing a 10-fold increase over the last 4 years. Analysis of such data has become a focus of much research within bioinformatics and computational biology because it is the SNPs that carry the information that distinguishes the individuals within a species. Encoded in this data is important information about the relationship between the characteristics of an individual and their genetic code. The investigators are developing methods and models for this data that are fundamentally different in approach from the standard techniques currently used in two areas where SNP data is used, coalescent and phylogenetic inference, in which one reconstructs genetic relationships similar to family trees based on the current ancestors only. The broader impacts of this project include the development of methods and models with potential wide ranging uses across broad scientific disciplines, the increased fusion of biological and mathematical innovation, and opportunities for broad interdisciplinary training of diverse students.