Mathematical and Statistical Models for Nucleosome

描述（申请人提供）：核小体是真核生物染色体中DNA与组蛋白八聚体结合形成的染色质的基本包装单位。核小体沿着基因组DNA的详细位置对于正确的基因调控至关重要--因此对于人类和所有其他真核生物的健康和发育至关重要--然而，目前，我们缺乏任何方法来“计算机模拟”预测基因组中核小体定位序列元件。该项目的长期目标是发展预测给定DNA序列的核小体形成倾向的能力，并将核小体形成序列的全基因组分布与染色体功能相关联。为达致这个目标，我们建议在未来数年逐步达致以下五个目标。(1)实验收集1000个酵母核小体核心DNA序列和1000个酵母双核小体DNA序列（通过接头DNA连接的两个相邻核小体）。 这些酵母核小体DNA序列将形成一个新的训练集以及一个验证集，用于酵母基因组中的核小体定位预测。(2)开发和完善统计方法，以比对核小体DNA序列。初步提出了一种新的核小体DNA序列比对统计模型。我们将通过引入基于自由能考虑的不同周期性双核苷酸信号的重要性的加权方案来改进该模型。(3)使用（2）中的比对来训练模型以预测核小体定位。考虑用非齐次马尔可夫链模型和“混合序列”模型来模拟核苷酸的序列依赖结构。(4)通过将相邻核小体的间距特性的知识并入模型来推广。由（1）产生的双核小体序列将提供核小体间空间的分布证据，并且dl 1极大地促进了对 全基因组核小体定位。(5)自动化开发的工具和算法。

DESCRIPTION (provided by applicant): The nucleosome is the fundamental packing unit of chromatin formed by association of DNA and histone octamer in eukaryotic chromosomes. The detailed locations of nucleosomes along genomic DNA are critical for proper gene regulation - and therefore for the health and development of humans and all other eukaryotes - yet, at present, we lack any methods for the "in-silico" prediction of nucleosome positioning sequence elements in genomes. The long-term goal of this project is to develop capability to predict nucleosome-forming propensity given DNA sequences and correlate the genome-wide distribution of nucleosome-forming sequences with chromosome function. Towards this goal, we propose to achieve the following five aims progressively in the subsequent years. (1) Experimentally collect 1000 yeast nucleosome core DNA sequences, and 1000 yeast di-nucleosome DNA sequences (two neighboring nucleosomes linked by linker DNA). These yeast nucleosome DNA sequences will form a new training set as well as a validating set for nucleosome positioning prediction across the yeast genome. (2) Develop and refine statistical methods to align nucleosome DNA sequences. A novel statistical model for nucleosorne DNA sequence alignment was proposed in the preliminary study. We will refine this model by introducing a weighting scheme based on the importance of different periodic di-nucleotide signals from free energy consideration. (3) Use the alignment in (2) to train a model to predict nucleosome positioning. An inhomogeneous Markov chain model and a "mixture train" model are under consideration to model the sequential dependent structure of nucleotides. (4) Generalize the model by incorporating knowledge in spacing properties of neighboring nucleosomes. The di-nucleosome sequences generated from (1) will provide distributional evidence for between-nucleosome space and dl1 greatly facilitate prediction of genome-wide nucleosome positioning. (5) Automate the developed tools and algorithms.