Model-based methods for single cell chromatin interactomic data

项目摘要/摘要 已在哺乳动物基因组中发现了数百万个顺式调节元件(Cre)，它们含有 很大一部分GWA型变异与复杂的人类疾病和特征有关。解读监管 Cre和Gwas变异体的靶基因仍然具有挑战性，因为大多数基因不仅仅是受调控的 由近一维(1D)附近的Cres。相反，Cres可以形成DNA环并调节 几百千个碱基的基因表达(S)。因此，对染色质空间有深刻的理解 组织可以阐明基因调控和疾病机制。在过去的十年里，染色质 构象捕获(3C)衍生技术(例如，原位Hi-C、捕获Hi-C、Chia-PET、Plac-Seq和 HiChIP)已被广泛用于提供染色质空间组织的全基因组视图。然而， 这些技术通常应用于块状组织或纯化的细胞系，不能揭示细胞类型的特异性 复杂组织中的染色质相互作用组。幸运的是，利用单细胞技术的力量， 单细胞Hi-C(Schi-C)和Schi-C衍生的多模式分析，包括单细胞甲基-HIC和单细胞Hi-C 核甲基-3C，在单细胞分辨率下研究染色质相互作用组， 为研究复杂组织和疾病相关细胞中染色质的空间组织提供了有力的工具 类型。虽然Schi-C实验技术取得了长足的进步，但计算方法 Schi-C数据的分析在很大程度上落后了。方法论上的差距主要体现在三个方面：(1) 现有的方法不能有效地提高稀疏Schi-C数据的分辨率。(2)方法少 消除每个单元格内Schi-C数据的系统性偏差，并针对批次效应进行调整 不同的细胞。(3)没有方法可用于检测Kb分辨率细胞类型特定的染色质相互作用 Schi-C数据。为了填补这些空白，我提出了主要的研究方向：(1)发展基于深度学习的 方法将稀疏染色质接触归因于每个细胞，(2)建立非参数回归模型 消除每个单元格内的系统性偏差，并调整不同单元格之间的批处理效果，(3)制定 基于全局和局部背景模型的混合方法识别特定细胞类型的染色质 相互作用，并预测与复杂人类疾病相关的GWA变异体的可能靶基因 特征，以及(4)开发独立的、用户友好的软件包来分析单细胞染色质 交互作用数据和传播结果。拟议研究的完成将为用户提供可靠的 友好的计算方法，使我们能够分析单细胞分辨率的3D基因组组织和 解释它们对基因表达和复杂人类疾病的调节作用。

PROJECT SUMMARY/ABSTRACT Millions of cis-regulatory elements (CRE) have been identified in mammalian genomes, which harbor large portion of GWAS variants associated with complex human diseases and traits. Interpreting the regulatory target genes of CRE and GWAS variants remains challenging, as majority of genes are not merely regulated by CREs in close one-dimensional (1D) vicinity. Instead, CREs can form DNA loops and regulate the expression of gene(s) from hundreds of kilobases (Kb) away. Thus, deep understanding of chromatin spatial organization can shed light on gene regulation and disease mechanisms. During the last decade, chromatin conformation capture (3C)-derived technologies (e.g., in situ Hi-C, capture Hi-C, ChIA-PET, PLAC-seq and HiChIP) have been widely used to provide a genome-wide view of chromatin spatial organization. However, these technologies are usually applied to bulk tissue or purified cell lines, and cannot reveal cell-type-specific chromatin interactome within complex tissues. Fortunately, harnessing the power of single cell technologies, single cell Hi-C (scHi-C) and scHi-C-derived multi-modal assays, including single cell Methyl-HiC and single- nucleus methyl-3C, have been rapidly advanced to study chromatin interactome at single cell resolution, providing powerful tools to study chromatin spatial organization in complex tissues and disease relevant cell types. While great strides have been made in scHi-C experimental technologies, computational methods for analyzing scHi-C data are largely lagging behind. The methodological gaps fall mainly in three aspects: (1) Current methods are inefficient to enhance resolution from extremely sparse scHi-C data. (2) Few methods exist for removing systematic biases of scHi-C data within each cell, and adjusting for batch effect across different cells. (3) No method is available to detect Kb resolution cell-type-specific chromatin interactions from scHi-C data. To fill in these gaps, I propose major research directions: (1) develop deep learning-based methods to impute sparse chromatin contacts in each cell, (2) develop non-parametric regression models to remove systematic biases within each cell, and to adjust batch effects across different cells, (3) develop a hybrid approach based on both global and local background models to identify cell-type-specific chromatin interactions, and predict putative target genes of GWAS variants associated with complex human diseases and traits, and (4) develop stand-alone, user-friendly software packages to analyze single cell chromatin interactomic data and disseminate results. Completion of the proposed study will provide robust and user friendly computational methods that allow us to analyze 3D genome organization at single cell resolution and interpret their regulatory role on gene expression and complex human diseases.