Statistical Methods for High-Resolution Multiscale Analysis in DNA Interactions

基因组生物学中的一个基本谜团是哺乳动物DNA序列（约2米长）的30亿个碱基对是如何折叠、成环和卷曲以适应直径约5-10微米的细胞核的。在过去的几年里，在推进我们对基因组如何在三维（3D）中折叠的理解方面取得了快速进展，主要是由测序技术的进步推动的。该项目的目标是开发数学模型，以更深入地了解3D基因组结构如何与健康发育中的基因表达相关联，以及这些折叠模式在疾病的发作和进展过程中如何出错。这个项目旨在通过染色质构象捕获实验揭示基因组折叠的组织原理。迄今为止，没有明确的最佳实践计算方法存在的比较基因组组织跨细胞类型或生物扰动。该项目的目的是开发数学模型和计算方法，以获得对遗传物质如何在不同细胞状态下折叠的新见解，并灵敏地检测这些折叠模式如何被药物，生长因子和基因组编辑等生物扰动动态改变。该项目的重点是开发方法，以灵敏地检测两大类3D染色质特征的动态变化：（1）亚巨链拓扑关联结构域显示出块结构，（2）两个遥远的基因组位点之间的精确的远程相互作用，导致插入基因组DNA的循环。参数和非参数规范化的方法来阐明这些功能将被探索和基准。这些功能的模型将被开发，导致扫描统计，以确定他们在规范化的3D接触地图。这些扫描统计的错误发现率控制的方法将开发基于异构泊松场的分析。

A fundamental mystery in genome biology is how the three billion base pairs of a mammalian DNA sequence (approximately 2 meters long) are folded, looped, and coiled to fit into a cell nucleus that is roughly 5-10 microns in diameter. Rapid progress has been made over the last few years in advancing our understanding of how the genome folds in three dimensions (3D), primarily driven by advances in sequencing technologies. The goal of this project is to develop mathematical models that will provide a deeper understanding of how 3D genome structure is connected to gene expression in healthy development, and how these folding patterns go awry during the onset and progression of disease. This project aims to shed new light into the organizing principles governing genome folding through chromatin conformation capture experiments. To date, no clear best practice computational methods exist for the comparison of genome organization across cell types or biological perturbations. The aim of this project is to develop mathematical models and computational methods to gain new insight into how the genetic material folds in different cellular states, and to sensitively detect how these folding patterns are dynamically altered by biological perturbations such as drugs, growth factors, and genome editing. This project focuses on developing methods to sensitively detect dynamic changes in two broad categories of 3D chromatin features: (1) sub-megabase topologically associating domains exhibiting a block structure, and (2) precise long-range interactions between two distant genomic loci, leading to looping out of the intervening genomic DNA. Both parametric and non-parametric normalization approaches for elucidating these features will be explored and benchmarked. Models for these features will be developed, leading to scan statistics for identifying them in normalized 3D contact maps. Methods for false discovery rate control for these scan statistics will be developed based on analysis of heterogeneous Poisson fields.