DETECTING SUBTLE SIGNALS IN GENOMIC SEQUENCE

据估计，人类基因组中约有80,000个基因(Fields C.，et al.(1994年)。为了将这一基因蓝图转变为功能有机体，基因必须以特定的时间和空间模式表达。寻找控制这种表达的信号并理解他们的语言是后基因组时代的主要挑战之一。实验室鉴定基因组序列中的调控元件、模块和区域通常是一个艰巨、耗时和昂贵的过程。如果能够开发出具体的方法，计算分析有望以最低的成本加速这一过程。这项拟议研究的长期目标是开发和应用贝叶斯生物信息学计算方法，该方法将描述基因组转录调控系统的完整接线图。这一描述将包括四个部分：1)识别所有转录因子超家族，并根据DNA识别基序和激活域将它们分类为功能相关的亚类；2)识别和鉴定基因组的转录调控模块及其内部的所有因子结合元件；3)全面描述因子与其结合元件之间的联系；4)描述可供选择的转录调节基序的特征，包括基于DNA组成、DNA和RNA结构的转录调控基序。这些目标将使用贝叶斯统计模型和算法来解决，这些模型和算法是我们在当前获奖期间开发的基础。其中包括组装转录因子超家族并对其进行多重比对的Gibbs采样算法、转录因子分类算法、用于描述组成和结构异质性、RNA二级结构和系统发生足迹的精确贝叶斯算法，以及用于调控模块识别和表征的递归Gibbs采样HMM。

It is estimated that there are approximately 80,000 genes in the human genome (Fields C., et al. 1994). To turn this genetic blueprint into a functional organism, genes must be expressed in a specific temporal and spatial pattern. Finding signals that control this expression and understanding their language is one of the major challenges of the post- genome era. Laboratory identification of regulatory elements, modules, and regions in genomic sequences is often an arduous, time-consuming, and expensive process. If specific approaches can be developed, computational analyses promise to accelerate this process at minimal cost. The long term goal of the proposed research is to develop and apply Bayesian bioinformatics computational methods which will describe the complete wiring diagram for a genome's transcription regulation system. This description will include four components: 1) the identification of all superfamilies of transcription factors and their classification into functionally related subclasses based on both the DNA recognition motifs and the activator domains; 2) the identification and characterization of a genome's transcriptional regulatory modules and all factor binding elements within them; 3) the full delineation of the connections between factors and their binding elements; 4) a characterization of alternative transcriptional regulatory motifs, including those based on DNA composition, and DNA and RNA structure. These goals will be addressed using Bayesian statistical models and algorithms, the foundations for which we developed during the current award period. These include Gibbs sampling algorithms to assembly superfamilies of transcription factors and multiply align them, transcription factor classification algorithms, exact Bayesian algorithms for the description of compositional and structural heterogeneity, RNA secondary structure, and phylogenetic footprinting, and recursive Gibbs sampling HMM for regulatory module identification and characterization.