Novel Statistical Inference for Biomedical Big Data

生物医学大数据的新颖统计推断

• 批准号: 10701041
• 负责人: ALI SHOJAIE
• 金额: $ 41.5万
• 依托单位: UNIVERSITY OF WASHINGTON
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-09-05 至 2024-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10701041
• 关键词: Address; Adoption; Behavioral; Big Data Methods; Biological; Biological Assay; Biological Markers; Code; Collaborations; Collection; Communities; Computer software; Data; Data Sources; Development; Dimensions; Disease; Electronic Health Record; Evaluation; Fostering; Galaxy; Genetic study; Goals; Heart; Imaging technology; Individual; Linear Models; Measurement; Measures; Medical Imaging; Methods; Modeling; Molecular; Multiomic Data; Outcome; Phenotype; Procedures; R programming language; Research Personnel; Sample Size; Scientist; Screening procedure; Software Tools; Structure; System; Technology; Testing; Trans-Omics for Precision Medicine; Uncertainty; Work; big biomedical data; computational pipelines; data integration; design; diverse data; effective therapy; experimental study; heterogenous data; high dimensionality; improved; interest; machine learning method; member; novel; open source; public health relevance; screening; simulation; statistical and machine learning; structured data; tool; treatment strategy; user friendly software

Project Summary This project develops novel statistical inference procedures for biomedical big data (BBD), including data from diverse omics platforms, various medical imaging technologies and electronic health records. Statistical inference, i.e., assess- ing uncertainty, statistical significance and confidence, is a key step in computational pipelines that aim to discover new disease mechanisms and develop effective treatments using BBD. However, the development of statistical inference procedures for BBD has lagged behind technological advances. In fact, while point estimation and variable selection procedures for BBD have matured over the past two decades, existing inference procedures are either limited to simple methods for marginal inference and/or lack the ability to integrate biomedical data across multiple studies and plat- forms. This paucity is, in large part, due to the challenges of statistical inference in high-dimensional models, where the number of features is considerably larger than the number of subjects in the study. Motivated by our team's extensive and complementary expertise in analyzing multi-omics data from heterogenous studies, including the TOPMed project on which multiple team members currently collaborate, the current proposal aims to address these challenges. The first aim of the project develops a novel inference procedure for conditional parameters in high-dimensional models based on dimension reduction, which facilitates seamless integration of external biological information, as well as biomedical data across multiple studies and platforms. To expand the application of this method to very high-dimensional models that arise in BBD applications, the second aim develops a data-adaptive screening procedure for selecting an optimal subset of relevant variables. The third aim develops a novel inference procedure for high-dimensional mixed linear models. This method expands the application domain of high-dimensional inference procedures to studies with longitu- dinal data and repeated measures, which arise commonly in biomedical applications. The fourth aim develops a novel data-driven procedure for controlling the false discovery rate (FDR), which facilitates the integration of evidence from multiple BBD sources, while minimizing the false negative rate (FNR) for optimal discovery. Upon evaluation using ex- tensive simulation experiments and application to multi-omics data from the TOPMed project, the last aim implements the proposed methods into easy-to-use open-source software tools leveraging the R programming language and the capabilities of the Galaxy workflow system, thus providing an expandable platform for further developments for BBD methods and tools.

项目摘要 该项目开发了生物医学大数据（BBD）的新型统计推断程序，包括来自潜水员的数据 OMICS平台，各种医学成像技术和电子健康记录。统计推断，即评估 - 不确定性，统计学意义和信心是计算管道的关键步骤，旨在发现新的 疾病机制并使用BBD开发有效的治疗方法。但是，统计推断的发展 BBD的程序落后于技术进步。实际上，虽然点估计和可变选择 BBD的程序在过去二十年中已经成熟，现有的推理程序要么仅限于简单 边际推断和/或缺乏在多个研究中整合生物医学数据的能力和/或 表格。在很大程度上，由于高维模型中统计推断的挑战，这种稀缺性在很大程度上是由于 特征数量比研究中的受试者数量大。由我们团队广泛的动机 以及来自异质研究的分析多摩学数据的完整专业知识，包括顶级项目 当前的提案旨在应对这些挑战。第一个 该项目的目的为基于高维模型的条件参数开发了一种新颖的推理程序 关于缩小尺寸，哪些设施无缝整合外部生物学信息以及生物医学 跨多个研究和平台的数据。将此方法的应用扩展到非常高的模型 在BBD应用程序中出现的第二个目的是开发一个数据自适应筛选程序，以选择最佳 相关变量的子集。第三个目标开发了高维混合线性的新型推理程序 型号。该方法将高维推理程序的应用领域扩展到具有纵向的研究 dinal数据和重复测量，通常在生物医学应用中产生。第四个目标发展了小说 控制错误发现率（FDR）的数据驱动程序，该程序主持了证据的整合 多个BBD来源，同时最大程度地降低了假阴性率（FNR）以进行最佳发现。在评估时使用Ex- 从顶级项目中进行的次级模拟实验和对多摩斯数据的应用，这是最后的目标实现 提出的方法用于易于使用的开源软件工具，利用R编程语言和 Galaxy Work流量系统的功能，从而为BBD提供了可扩展的平台 方法和工具。