Estimation and Inference of Gene Regulatory Networks

项目摘要 阿尔茨海默病(AD)是最常见的无法治愈的神经退行性疾病，大多数情况下 通常在疾病的晚期被诊断出来。为了增进我们对人类进化的起源、发展和发展的理解 以及这种疾病的病因学，许多遗传和基因组研究，包括全基因组关联研究 (Gwas)，成功地鉴定了一些相关的单核苷酸多态(SNPs)。 然而，由于这些相关SNP中的许多要么接近多个基因，要么远离任何基因， 识别致病基因，从而了解该病的生物学机制仍然是最重要的 很有挑战性。最重要的是解开和理解易感SNPs的调节作用 以及阿尔茨海默病等复杂人类疾病的基因，以便制定治疗和预防策略。 鉴于迫切需要了解这种疾病的生物学机制，私人投资机构建议开发 强大的统计和计算工具，用于准确推断AD患者和 健康的受试者。 拟议的项目由两个相互关联的部分组成：监管机构的重建和推理 基因和SNPs的网络。它将以监管网络结构为中心，做出特别努力 重点讨论了高维情况下发现和无偏推理的准确性，其中模型通过 描述监管网络的测距可能会大大超过样本量。关于重建 监管网络，该项目将开发约束最大似然方法来确定方向性 以及由基因之间的有向无环图(DAG)建模的成对因果关系的强度 和SNPs。关于网络推理，该项目将开发新的测试作为正式推理方法 达吉斯。在此基础上，将开发高维推理工具，用于检测多维结构的变化。 使用约束似然检验的样本网络。此外，还将研究线性和非线性因果关系。 在计算方面，将根据最先进的优化技术开发创新的战略，以及 用于大规模网络的可扩展分析。

Project Summary Alzheimer’s disease (AD) is the most common neurodegenerative disease without a cure, and most cases are often diagnosed in the late stage of the disease. To advance our understanding of the initiation, progression, and etiology of such a disease, many genetic and genomic studies, including genome-wide association studies (GWAS), have been conducted, successfully identifying some associated single-nucleotide polymorphisms (SNPs). However, since many of these associated SNPs are either close to multiple genes or far away from any genes, identifying the responsible genes and thus understanding biological mechanisms with the disease remain most challenging. Of paramount importance is unraveling and understanding the regulatory roles of susceptible SNPs and genes for complex human diseases such as AD, so that treatment and prevention strategies can be developed. Given the urgent need of understanding the biological mechanisms of the disease, the PIs propose to develop powerful statistical and computational tools for accurate inference of gene regulatory networks for AD patients and healthy subjects. The proposed project consists of two interconnected components: reconstruction and inference of regulatory networks of the genes and SNPs. It will be centered on structures of regulatory networks, with particular effort focused on the accuracy of discovery and unbiased inference in high-dimensional situations, where model pa- rameters describing regulatory networks may greatly exceed the sample size. With regard to reconstruction of regulatory networks, the project will develop the constrained maximum likelihood method to identify directionality as well as strengths of pairwise causal relations, modeled by directed acyclic graphs (DAGs) among the genes and SNPs. With regard to network inference, the project will develop novel tests as formal inference methods for DAGs. On this ground, high-dimensional inference tools will be developed for detecting structural changes of multi- ple networks using the constrained likelihood tests. Moreover, linear and nonlinear causal relations will be studied. Computationally, innovative strategies will be developed based on the state-of-the-art optimization techniques, and be used for scalable analysis of large-scale networks.