Sparse Structure Identification from High-Dimensional Epigenomic Data

描述（由申请人提供）：越来越多的证据支持组蛋白修饰的不同组合赋予不同功能特异性的假设。鉴定各种组蛋白修饰模式并将它们与基因组的功能元件联系起来是表观遗传学的一大兴趣。高通量实验技术，如ChIP-chip和ChIP-Seq，带来了丰富的组蛋白修饰数据。然而，目前的实验和计算方法只能在非常有限的程度上探索这些数据。该项目的长期目标是开发新的统计方法，用于从组蛋白修饰数据中识别稀疏结构。施加稀疏性是处理具有噪声信息和小样本量的极高维度数据的理想方法。提出了四个具体目标，包括：(1)鉴定基因组上的新功能位点；(2)不同监管要素之间的准确传播；(3)确定组蛋白修饰在调控中的相互作用；(4)揭示染色质特征的预测性DNA基序。为实现这些目标，将开发新的稀疏统计方法，包括结合变量选择的高维聚类方法、基于稀疏协方差估计的降维分类方法、多功能元素图形模型的联合估计以及多响应多预测回归方法。这个项目将由两名统计学家和一名生物化学家合作进行。所提出的方法将通过已发表的数据集和表观基因组路线图项目提供的数据集进行验证并应用于其中一个pi所涉及的数据集。

DESCRIPTION (provided by applicant): Evidence is accumulating to support the hypothesis that different combinations of histone modifications confer different functional specificities. Identification of various histone modification patterns and linking them with functional elements of the genome is of great interest in epigenetics. High-throughput experimental techniques, such as ChIP-chip and ChIP-Seq, lead to a rich amount of histone modification data. However, current experimental and computational methods have only been able to explore these data to a very limited extent. This project bears a long-term objective of developing novel statistical methods for sparse structure identification from histone modification data. Imposing sparsity is an ideal way for handling extremely high-dimensional data with noisy information and small sample size. Four specific aims are proposed, including (1) identification of new functional sites on the genome; (2) accurate dissemination between different regulatory elements; (3) identification of the interaction between histone modifications in regulation; (4) uncovering the predictive DNA motifs of the chromatin signature. Novel sparse statistical methods will be developed to achieve these aims, including a high-dimensional clustering method combined with variable selection, a classification method featured by sparse covariance estimation based dimension reduction, a joint estimation of graphical models for multiple functional elements, and a multi-response multi-predictor regression method. This project will be conducted through the collaboration between two statisticians and a biochemist. The proposed methods will be validated through and applied to both published datasets and those provided by the epigenome roadmap project in which one of the PIs is involved.