Statistical Tools for Whole-Genome Analysis & Prediction of Complex Traits and Diseases

 描述（由申请人提供）：大基因组数据的分析需要专门的软件，能够科普数据本身的高维性质和潜在生物机制的复杂性带来的挑战。在NIH的支持下，我们开发，测试并维护了贝叶斯广义线性回归R库（可在CRAN，BGLR，Pérez和de los Campos 2014中获得）：一个全面的贝叶斯统计软件，实现了大量的全基因组回归（WGR）程序，包括线性模型和半参数回归（RKHS）的收缩和变量选择方法。几项使用BGLR分析大型基因组数据集（具有数十万个SNP和数千个个体）以及多层组学数据的研究证明了该软件的价值。为了更新我们的资助，我们提出了一系列改进和发展，使BGLR更适合大数据分析，并将大大扩展所实施的模型类别。我们将制定和实施：（目标1）使BGLR能够使用存储在分布式二进制文件中的输入进行计算的方法，而无需将数据完全加载到RAM中-这将为大组学数据集的分析打开巨大的机会;（目标2）BGLR模块，以适应各种各样的交互模型，包括分类（例如，性别、治疗）或定量（例如，BMI）风险因素与全基因组数据（例如，SNP，表达谱）;（目标3）并入先前信息的方法（例如，注释）到全基因组回归;和，（目标4）在线培训的工具。我们目标的成功实现将为研究人员提供高效的数据分析工具，用于大型组学数据集的全基因组分析。

 DESCRIPTION (provided by applicant): The analysis of big genomic data requires specialized software able to cope with challenges emerging from both the high dimensional nature of the data itself and the complexity of the underlying biological mechanisms. With NIH support we developed, tested and now maintain the Bayesian Generalized Linear Regression R-library (available at CRAN, BGLR, Pérez and de los Campos 2014): a comprehensive Bayesian statistical software that implements a large collection of Whole-Genome Regression (WGR) procedures, including shrinkage and variable selection methods for linear models and semi parametric regressions (RKHS). Several studies that have used BGLR for analyses of large genomic data sets (with hundreds of thousands of SNPs and thousands of individuals) as well as multi-layer omic data demonstrate the value of the software. For the renewal of our grant we propose a set of improvements and developments that will make BGLR better suited for the analysis of Big Data and will greatly expand the classes of models implemented. We will develop and implement: (Aim 1) methods to enable BGLR to carry out computations using inputs that are stored in distributed binary files, without fully loading data into RAM-this will open great opportunities for the analysis of big omic data sets; (Aim 2) a BGLR module to fit a diverse array of interaction models, including interactions between categorical (e.g., sex, treatment) or quantitative (e.g., BMI) risk factors with whole- genome data (e.g., SNPs, expression profiles); (Aim 3) methods to incorporate prior information (e.g., annotation) into whole genome regressions; and, (Aim 4) instruments for online training. The successful achievement of our aims will provide researchers with efficient data analysis tools for whole-genome analysis of large omic data sets.