High dimensional statistical data modeling and integration for studying regulatory variation

项目摘要 哺乳动物细胞的基因调控程序在很大程度上受到长距离转录因子的影响。 染色质相互作用我们建议开发鲁棒性和可扩展的统计方法 对于两个关键的基因组推理问题， 交互.首先，在单细胞水平上与3C- 基于scHi-C的方法是充分理解细胞类型特异性基因的基础 调控scHi-C测量结果包含未探索的生物多样性。但这些 测量结果易于出现极端稀疏、技术偏差和噪声。虽然最初的 推断方法简单地集中于scHi-C数据的低维表示， 缺乏可以利用数据去噪中的非线性的可扩展框架 阻碍了这些实验的关键推理任务。我们将解决这些关键问题， 通过为scHi-C数据开发新的深度生成模型来克服这些缺点。由- 这些方法将提高感兴趣的信号的功率， 被研究。第二，虽然在测序和大规模可用性方面取得了进展， 表观基因组数据提高了全基因组关联的能力和解释 研究（GWAS），在确定哪些基因非编码SNP可能是 通过长距离染色质相互作用影响阻碍GWAS的翻译 临床干预的发现。利用现有的大规模多样性研究 远交小鼠，我们将开发一个严格的框架，整合多组学功能 用于精细绘制模式生物分子数量性状基因座和转移的数据模式 将非编码GWAS SNP与它们的效应子连接的结果提供给人类，即， 易感性基因2型糖尿病（T2 D）的大规模应用将提供 候选T2 D效应基因和它们的调节基因座， 实验性后续。这两个目标将通过以下结合来实现： 方法学发展，理论分析，数据驱动模拟，计算 分析和实验验证。本项目产生的统计资源 将作为开源软件传播。该项目的成功完成将 有助于确保从强大scHi-C实验中获得最大信息 和模式生物多组学数据。

Project Summary Gene regulatory programs of mammalian cells are largely influenced by long-range chromatin interactions. We propose to develop robust and scalable statistical methods for two critical genomic inference problems hinging upon long-range chromatin interactions. First, the study of long-range interactions at the single cell-level with 3C- based method scHi-C is fundamental to fully understanding cell type-specific gene regulation. scHi-C measurements harbor unexplored biological diversity. However, these measurements are prone to extreme sparsity, technological bias, and noise. While initial inference methods simply focused on lower dimensional representations of scHi-C data, lack of a scalable framework that can exploit nonlinearities in de-noising of the data impedes key inference tasks from these experiments. We will address these critical shortcomings by developing a novel deep generative model for scHi-C data. By de- noising the data, these methods will improve the power with which signals of interest can be studied. Second, while advances in sequencing and large-scale availability of epigenome data improved the power and interpretation of genome-wide association studies (GWAS), shortcomings in identifying which genes noncoding SNPs might be impacting through long-range chromatin interactions hinder the translation of GWAS findings into clinical interventions. Leveraging existing large-scale studies of diversity outbred mice, we will develop a rigorous framework that integrates multi-omics functional data modalities to fine-map model organism molecular quantitative trait loci and transfer the results to humans for linking noncoding GWAS SNPs to their effector, i.e., susceptibility, genes. Large-scale application with type 2 diabetes (T2D) traits will deliver candidate T2D effector genes and their regulatory loci that are amenable for experimental follow-up. Both aims will be accomplished through a combination of methodological development, theoretical analysis, data-driven simulation, computational analysis, and experimental validation. Statistical resources generated from this project will be disseminated as open-source software. Successful completion of the project will help to ensure that maximal information is obtained from powerful scHi-C experiments and model organism multi-omics data.