Methods and Theory for Estimating Individual-Specific and Cell-Type-Specific Gene Networks

在生物网络分析的现有文献中，大多数方法假设跨学科的共同或分层网络结构。因此，他们无法灵活地解释个体层次网络的异质性。例如，由于遗传变异、性别和生物样本中不同成分的复杂影响，个体水平的网络可能会有所不同。这种网络异质性的特点提出了迫切需要新的统计方法和理论。受基因共表达分析的启发，本项目的目标是在异质性网络分析方面取得实质性进展。所开发的方法可以影响人类遗传学和基因组学、精准健康和医学的广泛主题;它们也更广泛地适用于神经科学、金融和社会科学等科学领域。PI将通过培训本科生和研究生以及开发专题课程将研究融入教育。该项目旨在为估计具有异质性的高维协方差的新挑战提供新颖和基本的视角。该项目的第一部分在估计个人特定的图形模型方面开辟了新的天地。PI将开发一个新的图形回归模型，该模型将条件依赖结构与高维协变量联系起来。第二部分解决了从具有不同组成的聚合数据推断细胞类型特异性基因网络的挑战。PI将开发一个灵活的框架，该框架不对表达式的分布做出特定假设，并考虑一种新的最小二乘估计。在这个项目中开发的方法具有吸引人的特点，包括可识别性和可解释性，有效的计算，可量化的统计误差，和有效的统计推断。这个奖项反映了NSF的法定使命，并已被认为是值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估的支持。

In the existing literature on biological network analyses, most approaches assume a common or stratified network structure across subjects. Consequently, they are not able to flexibly account for heterogeneity in individual-level networks. For example, the individual-level networks may differ due to the complex effects of genetic variants, sex, and varying compositions across biological samples. Characterizing such network heterogeneity presents an urgent need for new statistical methodology and theory. Motivated by gene co-expression analyses, this project aims to make substantial progress in network analysis with heterogeneity. The developed methods can impact a wide range of topics in human genetics and genomics, precision health, and medicine; they are also more broadly applicable to scientific fields such as neuroscience, finance, and social science. The PI will integrate research into education by training undergraduate and graduate students and developing special topics courses.This project aims to provide novel and fundamental perspectives on the emerging challenges in estimating high-dimensional covariances with heterogeneity. The first part of the project breaks new ground on estimating individual-specific graphical models. The PI will develop a new graphical regression model that relates the conditional dependence structure to covariates of high dimensions. The second part addresses the challenge in inferring cell-type-specific gene networks from aggregated data with different compositions. The PI will develop a flexible framework that does not make specific assumptions on the distributions of expressions and consider a novel least squares estimation. The developed methods in this project have appealing features, including identifiability and interpretability, efficient computation, quantifiable statistical errors, and valid statistical inference.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.