Statistical methods for data integration

前言：基因组学研究和应用的性质在过去二十年中发生了巨大变化。越来越便宜的新技术正被用来产生大量的生物数据，目的是获得对分子生物学的基本了解，阐明疾病的病因，并评估基因组信息在临床医学和公共卫生领域的潜在预测效用。正在生成的各种基因组数据集可能包含独立和冗余信息。高通量生物数据也往往是嘈杂的。假设：新的稀疏统计方法可以捕捉噪声数据的基本特征，综合统计分析允许跨数据集合成信息，将提高我们可靠估计效应大小的能力，在检测关联方面提供更多的力量，并导致更准确的预测。此外，通过在综合分析中考虑相对重要的措施，将获得更好的结果。* 目标1：异构数据的集成方法。我们将开发基于核的统计方法，用于监督和非监督集成。我们将为集成异构数据类型提供一个统一的概念和方法框架。这将包括将基因组数据与环境、临床和实验室测量相结合的方法。我们将使用模拟来比较和对比方法，并使用我们合作者的数据以及公共领域的数据进行经验评估。目标2：用于微生物组数据分析的综合多变量方法 * 我们将开发用于微生物组数据分析的多变量方法，该方法包含稀疏表示并扩展综合框架以过滤掉噪声特征。最后，我们将开发联合分析多种相关表型的方法。*重要性：使用多管齐下的综合方法和适当的稀疏统计方法，将产生关键信息，可以确定遗传和临床变异性的准确评估，并将最大限度地提高结局的准确和有效评估。高素质的人员（HQP）将接受培训，基于我们工作的软件将免费提供给更广泛的科学界。

Preamble: The nature of genomics research and application has changed drastically in the last two decades.***Increasingly inexpensive new technologies are being used to produce large biological data with the aim of gaining fundamental understanding of molecular biology, elucidating etiology of diseases and assessing the potential predictive utility of genomic information in clinical medicine as well as public health areas. The various genomic data sets that are being generated may contain both independent and redundant information. High throughput biological data also tend to be noisy.******Hypothesis: Novel sparse statistical methods that can capture essential characteristics of noisy data, and integrative statistical analyses that allow synthesis of information across data sets, will improve our ability to estimate effect sizes reliably, provide more power in detecting associations and lead to more accurate predictions. Furthermore, improved results will be obtained by considering measures of relative importance in the integrative analyses.*** ***AIM 1: Integrative Methods for Heterogeneous Data. We will develop kernel based statistical methods for supervised and unsupervised integration. We will provide a unified conceptual and methodological framework for integrating heterogeneous data types. This will include approaches for integrating genomic data with environmental, clinical and laboratory measurements. We will compare and contrast methods using simulations, and empirically evaluate using data from our collaborators as well as data available in the public domain.***AIM 2: Integrative Multivariate Methods for the Analysis of Microbiome Data***We will develop multivariate methods for the analysis of microbiome data that incorporate sparse representation and extend an integrative framework to filter out noisy features. Finally we will develop methods for jointly analyzing multiple correlated phenotypes.***Significance: Using multi-pronged integrative approaches and with appropriate sparse statistical methodologies, critical information will be generated, accurate assessment of genetic and clinical variability can be determined, and precise and valid assessment of outcomes will be maximized. Highly qualified personnel (HQPs) will be trained and software based on our work will be made freely available to the wider scientific community.**