Decoding Gene Expression Control Using Conditional Clustering by Dynamics

描述（由申请人提供）：微阵列技术使研究人员能够同时测量数千个基因的表达，并有望为基因组的调控机制带来新的曙光。利用这一技术进行实验研究的主要途径是基于细胞机制的时间解剖。时间实验提供了观察这些作用机制的可能性，并将基因组分解为参与相同过程的基因组。该项目的总体目标是开发一种无监督的方法和集成的软件环境，以自动发现时序微阵列实验的调节机制。支持我们方法的假设是，复杂的相互作用模式可以通过分析条件而不是边缘基因表达谱来识别。这种新颖的方法也为实验设计和抽样策略提供了原则性的指导，并且它自然地扩展到一大类统计模型，能够捕获更广泛的动态行为和实验设计。我们计划开发一个全面的框架来设计和分析通过时间实验收集的微阵列数据。该框架将用于指定和回答样本大小和采样频率确定的关键设计问题。利用这个框架，我们将开发一种新的基于模型的方法和迭代搜索算法，称为条件聚类，通过分析给定一组其他基因的基因行为，而不是孤立地分析每个基因的行为，来识别由一组基因决定的不同行为模式。我们将在一个计算机程序中实现这个设计和分析框架，该程序将在互联网上发布。该项目汇集了人工智能，理论统计和实验设计方面的研究人员，他们在生物信息学方面有长期的方法论贡献，以开发一种新的方法方法来解决基因组研究前沿的关键问题。

DESCRIPTION (provided by applicant): Microarray technology enables investigators to simultaneously measure the expression of thousands of genes and holds the promise to cast new light onto the regulatory mechanisms of the genome. A main avenue of experimental investigation, leveraging on this technology, is based on the temporal dissection of cellular mechanisms. Temporal experiments offer the possibility of observing these mechanisms in action and to break down the genome into sets of genes involved in the same processes. The overall goal of this project is to develop an unsupervised approach and an integrated software environment to automatically discover regulatory mechanisms from temporal microarray experiments. The hypothesis underpinning our approach is that complex interaction patterns can be identified through analysis of conditional rather than marginal gene expression profiles. This novel approach also provides principled guidance to experimental design and sampling strategies, and it naturally extends to a large class of statistical models, able to capture a wider range of dynamic behaviors and experimental designs. We plan to develop a comprehensive framework to design and analyze microarray data collected through temporal experiments. This framework will be used to specify and answer the critical design questions of sample size and sampling frequency determination. Using this framework, we will develop a new model-based approach and an iterative search algorithm, called Conditional Clustering, to identify different patterns of behavior determined by a set of genes through the analysis of the behavior of a gene given a set of other genes, rather than the behavior of each gene in isolation. We will implement this design and analysis framework in a computer program that will be distributed over the Internet.This project brings together researchers in artificial intelligence, theoretical statistics and experimental design with a long track record of methodological contributions to bioinformatics to develop a novel methodological approach to a critical question at the forefront of genomic research.