Sparse and Efficient Estimation with Semiparametric Models in Meta-Analysis

在许多科学领域，如基因组学、流行病学和经济学，将多个研究的大规模数据集结合起来是充分利用收集到的数据的一种有价值的方法。然而，此类研究通常有隐私政策，以防止个人层面的数据共享。例如，在生物医学研究中，虽然数据整合可以提高评估疾病风险因素的能力，但研究协议通常禁止在研究之间共享参与者水平的基因组和临床数据。该项目将研究综合分析，在汇总统计中使用压缩信息，而不需要个人数据。PI计划建立一个具有半参数回归模型的广泛框架，并开发具有理论保证的具体和计算效率的方法。本科生和研究生都将通过参与研究项目接受培训。PI将研究半参数回归元分析的一般似然理论。要建立的理论框架将包括对产生各种数据类型的不同观察方案的研究的元分析。该项目将处理元分析的同质和异质结构。PI将开发基于汇总统计的半参数方法，目的是有效估计和稀疏结构恢复。在开发方法的过程中，研究将集中在使用和扩展最小二乘近似和正则化等技术。本文将在两种结构类型的元分析下研究这些方法的统计性质和优化算法。由此产生的方法将适用于各种研究，如大规模公共卫生研究、预后特征研究和全基因组关联研究。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

In many scientific fields, such as genomics, epidemiology, and economics, combining large-scale datasets of multiple studies is a valuable approach to fully utilizing the collected data. However, such studies often have privacy policies that prevent individual-level data sharing. In biomedical research, for example, while data integration can boost the power of evaluating risk factors of a disease, study protocols typically prohibit sharing participant-level genomic and clinical data among the studies. This project will investigate meta-analysis that combines studies using compressed information in summary statistics without requiring individual-level data. The PI plans to establish a broad framework with semiparametric regression models and to develop concrete and computationally efficient methods with theoretical guarantees. Both undergraduate and graduate students will receive training through involvement in the research project.The PI will study the general likelihood theory for meta-analysis with semiparametric regression. The theoretical framework to be established will embrace meta-analysis of studies with different observation schemes that generate various data types. The project will deal with both homogeneous and heterogeneous structures of meta-analysis. The PI will develop semiparametric methods based on summary statistics with the aim of efficient estimation and sparse structure recovery. In developing the methods, the research will focus on using and extending techniques such as least-squares approximation and regularization. Statistical properties and optimization algorithms of these methods will be studied under both structure types of meta-analysis. The resulting methods will be applicable to various studies such as large-scale public health studies, prognostic signature studies, and genome-wide association studies.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.