Statistical and Computational Tools for Next-generation ChIP-seq Applications

描述(申请人提供)：CHIP-SEQ是一种强大的绘制全基因组蛋白质-DNA相互作用(PDI)的技术。它越来越多地被世界各地的科学家用来研究正常细胞中的基因活动是如何受到控制的，以及为什么它们在疾病中被破坏。将CHIP-SEQ应用于基因调控研究面临着三大挑战：(1)如何分析大量的CHIP-SEQ数据集，以发现不同生物背景下基因调控的动态变化；(2)如何在对所有转录因子(TF)进行CHIP-SEQ不可行的实际约束下推断全局调控方案；(3)如何在杂合子SNPs的少量数据导致统计能力较低的情况下分析等位基因特异性事件。这项研究调查了新的统计和计算解决方案 以应对上述挑战。首先，将开发一种新的方法来发现和表征不同生物背景下基因调控的动态变化。广义差分主成分分析(GPCA/GDPCA)将无监督模式发现、降维和统计推断集成到一个统计框架中。它提供了一个系统的解决方案来分析涉及多个蛋白质的大型CHIP-SEQ数据集中的定量和曲线形状变化。预计它将有广泛的应用。其次，将开发一个计算框架来预测全球基因调控动态，即所有DNA结合基序信息可用的转录因子下游调控事件的动态变化。该分析将组蛋白修饰芯片SEQ、DNase-SEQ和FIRE-SEQ数据与DNA序列、公共芯片SEQ和公共基因表达数据的动态变化相结合。它将为利用CHIP-SEQ同时研究多个TF提供一个实用的、负担得起的、相当准确的解决方案。还将进行一项系统的基准研究，以评估技术、数据类型和分析方法对预测性能的影响。这项基准研究将为未来设计信息丰富的实验提供指导。第三，将开发一种检测等位基因特异性蛋白-DNA结合(ASB)的方法。该方法能够集成来自多个芯片序列数据集和完全阶段性基因组序列的信息，从而显著提高ASB推断的统计能力。各种偏见的来源也将得到处理。这项研究产生的指南和新的分析工具将使人们能够在未来设计信息丰富的CHIP-SEQ实验，这样通过收集一组CHIP-SEQ数据，人们不仅可以识别PDI的位置，还可以推断不同生物背景下TF结合位点的全球动态变化，如果有基因型数据，就可以有力地分析等位基因特异的基因调控。这将使CHIP-SEQ成为一个具有多种用途的低成本、高回报的实验。通过显著扩展CHIP-SEQ的实用性和增强能力，我们的计算基础设施有望对推进未来对基因调控的研究和剖析人类疾病背后的调控机制产生重大影响。

DESCRIPTION (provided by applicant): ChIP-seq is a powerful technology to map genome-wide protein-DNA interactions (PDIs). It is increasingly used by scientists worldwide to study how gene activities are controlled in normal cells and why they are disrupted in diseases. Applying ChIP-seq to study gene regulation faces three major challenges: (1) how to analyze large ChIP-seq data sets to discover dynamic changes of gene regulation across different biological contexts, (2) how to infer global regulatory programs under the practical constraint that it is not feasible to conduct ChIP-seq for all transcription factors (TFs), and (3) how to analyze allele-specific events given the small amount of data at heterozygote SNPs which cause low statistical power. This study investigates novel statistical and computational solutions to address the challenges above. First, a new method will be developed to discover and characterize dynamic changes of gene regulation across different biological contexts. This method, Generalized Differential Principal Component Analysis (dPCA/GDPCA), integrates unsupervised pattern discovery, dimension reduction and statistical inference into a single statistical framework. It provides a systematic solution to analyze quantitative and curve shape changes in large ChIP-seq data sets involving multiple proteins. It is expected to have a wide range of applications. Second, a computational framework will be developed to predict global gene regulation dynamics, i.e., dynamic changes of downstream regulatory events of all TFs for which DNA binding motif information is available. The analysis integrates the dynamic changes of histone modification ChIP-seq, DNase-seq, and FAIRE-seq data with DNA sequences, public ChIP-seq, and public gene expression data. It will provide a practical, affordable, and reasonably accurate solution to utilizing ChIP-seq to study many TFs simultaneously. A systematic benchmark study will also be con- ducted to evaluate the impact of technologies, data types and analytical methods on prediction performance. This benchmark study will provide guidelines for designing informative future experiments. Third, a method for detecting allele-specific protein-DNA binding (ASB) will be developed. The method is able to integrate information from multiple ChIP-seq data sets and completely phased genome sequences to significantly improve the statistical power of ASB inference. Various sources of biases will also be handled. Guidelines and new analytical tools generated by this study will allow one to design informative ChIP-seq experiments in the future such that by collecting one set of ChIP-seq data, one can not only identify locations of PDIs, but also infer global dynamic changes of TF binding sites across different biological contexts, and, if genotype data are available, robustly analyze allele-specific gene regulation. This will make ChIP-seq a low-cost high-reward experiment that serves multiple purposes. By significantly expanding the utility and increasing the power of ChIP-seq, our computational infrastructure is expected to have a major impact on advancing future studies of gene regulation and dissections of regulatory mechanisms behind human diseases.