Novel Statistical Methods with Application to Imaging Genetics

成像遗传学是一个快速发展的领域，它将个人医学图像与遗传信息相结合，以评估遗传变异对大脑功能和结构的影响。这项建议的长期目标是为成像遗传学研究开发新的统计方法。随着先进的成像和基因分型技术的出现，已经进行了许多大规模的生物医学研究来收集成像和遗传数据。传统的成像遗传学分析方法存在计算量大、算法不稳定等问题，这就需要发展快速有效的统计方法。这一建议不仅将促进统计成像遗传学分析的方法学和理论的发展，而且将促进人们对成像和遗传数据之间关系的理解。拟议的项目有以下相互关联的主题。首先，PI将高维遗传数据作为标量预测器，将复杂的成像数据作为张量响应，从而建立高维张量响应回归模型。将研究高维张量响应线性回归、分位数回归和半参数回归模型。其次，PI将开发一个高维功能反应线性回归模型，并在两个不同的方向上开发统计推断程序。对于上述模型，推理程序仅适用于常见的变异基因数据。然后，PI将研究稀有变种存在时的情况，并开发适用于此情况的方法。*本提案中开发的统计方法是及时和重要的，将与许多大型真实数据集相关，例如阿尔茨海默病神经成像倡议、人类连接组项目、英国生物库数据、儿科成像、神经认知以及遗传学和费城神经发育队列数据。将为本科生和研究生以及博士后研究人员提供出色的培训，帮助他们获得宝贵的技能，增加他们在就业市场上的机会或开始他们的学术生涯。为了便于使用拟议的新方法，PI将在R或MatLab中实现这些方法，并向公众提供软件，同时发布相应的研究报告。

Imaging genetics is a rapidly evolving field that integrates individual medical images with genetic information to assess the impact of genetic variation on brain function and structure. The long term objective of this proposal is to develop novel statistical methods for imaging genetics studies. With the advent of both advanced imaging as well as genotyping techniques, many large biomedical studies have been conducted to collect imaging and genetic data. Conventional approaches for imaging genetics analysis suffer from issues such as heavy computational cost and algorithmic instability, which requires the development of fast and efficient statistical methods.******This proposal will not only enhance methodological and theoretical developments for statistical imaging genetics analysis, but also advance the understanding of the relationship between imaging and genetic data. The proposed project has the following interrelated themes. First, the PI will develop high dimensional tensor response regression models by treating the high dimensional genetic data as scalar predictors and the complex imaging data as a tensor response. The high dimensional tensor response linear regression, quantile regression and semiparametric regression models will be studied. Second, the PI will develop a high dimensional functional response linear regression model, and develop statistical inference procedures in two different directions. For the aforementioned models, the inference procedures only work for the common variants genetic data. The PI will then study the scenario when the rare variants exist, and develop methods that work for this scenario. ******The statistical methods developed in this proposal are timely and important and will be relevant to many large-scale real data sets, for example the Alzheimer's Disease Neuroimaging Initiative, the Human Connectome Project, the UK Biobank data, the Pediatric Imaging, Neurocognition, and the Genetics and Philadelphia Neurodevelopmental Cohort data. Undergraduate and graduate students as well as postdoctoral researchers will be provided with excellent training to help them to gain valuable skills that increase their chances in the job market or to start their academic career. In order to facilitate the use of the proposed new methods, the PI will implement them in R or Matlab and make software available to the public, along with publishing the corresponding research reports.