Statistical modeling of long-range chromatin interactions on gene regulation and underlying molecular

已发现与疾病相关的遗传变异在基因表达的调控基因组区域中丰富。为了详细了解疾病的发病机制，一个中心问题是系统地描述组织特异性基因表达程序是如何调控的。增强子是一个重要的调控元件家族，具有复杂的信号，在人类基因组中含量丰富。增强子对基因表达的调控是高度组织特异性的，与组合转录因子结合有关，并涉及远距离的三维染色质相互作用。因此，描述大规模的增强子调控网络是一项具有挑战性的工作。该项目的主要目标是开发一套概率模型和高效的机器学习算法，以预测不同细胞背景下的基因组范围的增强子调控网络和相关的分子机制，以建立远程相互作用，这将被用来解释与疾病相关的遗传变异。在目标1中，将开发新的综合图形模型来预测将组织特异性增强子与其远端靶基因连接起来的长范围染色质相互作用，以及组合转录因子结合模式。在目标2中，将设计计算算法来询问特定的染色质相互作用是如何建立的，从而导致对染色质形成的机械性见解。在目标3中，将开发统计模型，将增强基因调控网络与遗传学数据相结合，以预测哪些非编码变异可能扰乱调控链接并导致疾病，从而提高统计能力和准确性。这个建模框架将大大扩展对非编码变体和人类疾病机制的分析能力。来自这三个目标的计算预测将在乳腺癌发展的小鼠模型中进行实验测试。该项目将带来创新的计算工具和系统的生物学洞察，了解长程增强子调节及其在人类疾病中的功能作用。

Disease-associated genetic variants have been found to be enriched in regulatory genomic regions of gene expression. In order to gain detailed understanding of disease mechanisms, one central question is to systematically delineate how tissue-specific gene expression programs are regulated. Enhancers are a major family of regulatory elements with complex signatures and they are abundant in the human genome. Enhancer regulation of gene expression is highly tissue-specific, associated with combinatorial transcription factor binding, and involved with long-range three-dimensional chromatin interactions. It is therefore challenging to characterize the large-scale enhancer regulatory networks. The primary goal of this project is to develop a suite of probabilistic models and efficient machine learning algorithms to predict genome-wide enhancer regulatory networks in diverse panels of cellular contexts and the associated molecular mechanisms to establish long-range interactions, which will be leveraged to interpret disease-associated genetic variants. In Aim 1, novel integrative graphical models will be developed to predict long-range chromatin interactions linking tissue-specific enhancers to their distal target genes, along with combinatorial transcription factor binding patterns. In Aim 2, computational algorithms will be designed to interrogate how specific chromatin interactions are established, leading to mechanistic insights on chromatin formation. In Aim 3, statistical models will be developed to integrate enhancer-gene regulatory networks with genetics data to predict which non-coding variants may disrupt regulatory links and cause diseases, with improved statistical power and accuracy. This modeling framework will substantially expand the analytical ability on non-coding variants and human disease mechanisms. Computational predictions from the three aims will be experimentally tested in mouse models of breast cancer development. This project will lead to both innovative computational tools and systematic biological insights on long-range enhancer regulation and their functional roles in human diseases.