Study of dimension reduction methods driven by large scale biological data

大规模生物数据驱动的降维方法研究

• 批准号: 0707160
• 负责人: Ker-Chau Li
• 金额: $ 14万
• 依托单位: University of California-Los Angeles
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2007
• 资助国家: 美国
• 起止时间: 2007-09-01 至 2011-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0707160&HistoricalAwards=false
• 关键词: Study; dimension; reduction; methods; driven

The growing awareness of the importance of dimension reduction in bioinformatics has led to the development of new techniques. Various extensions of sliced inverse regression(SIR) have appeared. More recently a new way of studying the relationship between variables in a complex data system, called liquid association (LA), was proposed. LA describes how variation in the pattern of association between a pair of variables, including its sign and strength, is mediated by a third variable from the background. In this proposal, the investigators will (1) develop hybrid dimension reduction methods that blend in ideas from PCA (principal component analysis), K-mean, SIR, PHD (Principale Hessian Direction), LA and a variety of clustering methods; (2) provide effective dimension reduction tools for incorporating clinical or other phenotype data involving life time which are usually subject to censoring; (3) investigate related statistical inference issues concerning false positives; (4) investigate the pattern of cellular coordination at the functional module level; (5) develop LA based methods for variable selection; (6) provide better insight to intrinsic nonlinearity in marginal Gaussian models.The recent rapid growth in the public repertoire of biological data and knowledge resources has been astonishing. This includes the completion of genome sequencing for human and many species, the stride in the SNP detection and international HapMap project, the accumulation of full genome microarray gene expression data under a number of conditions for numerous organisms and tissues, the identification of the high density genetic markers, protein-protein interaction and complexes, as well as the availability of various gene annotation websites featuring both functional and structural information of the gene products, their biological roles and relevance to disease studies. Such open data resources hold the promise of benefiting numerous projects aiming at solving detail genetic profiles predisposing to complex diseases and their trait components. However, researchers are also facing the insurmountable difficulties due to the enormous complexity of data structure and exceedingly high dimensionality. To succeed, it is critical to continue the research on developing new computational arsenals. Modeling through probabilistic and statistical reasoning has found numerous compelling applications.

降维在生物信息学中的重要性的日益认识导致了新技术的发展。切片逆回归（SIR）的各种扩展已经出现。最近提出了一种研究复杂数据系统中变量之间关系的新方法，称为液体关联（LA）。LA描述了一对变量之间的关联模式的变化，包括它的符号和强度，是如何由背景中的第三个变量介导的。在这项提案中，研究人员将（1）开发混合降维方法，融合PCA的思想（主成分分析），K均值，SIR，PHD（Principale Hessian Direction）、LA和各种聚类方法;（2）提供有效的降维工具，用于合并涉及寿命的临床或其他表型数据，这些数据通常受到删失;（3）研究与假阳性相关的统计推断问题;（4）研究功能模块水平上的细胞协调模式;（5）开发基于LA的变量选择方法;（六）提供了更好的洞察边际高斯模型的内在非线性。最近的快速增长，在公共剧目的生物数据和知识资源惊人。这包括人类和许多物种基因组测序的完成，SNP检测和国际HapMap计划的跨越，许多生物和组织在许多条件下的全基因组微阵列基因表达数据的积累，高密度遗传标记，蛋白质-蛋白质相互作用和复合物的鉴定，以及提供各种基因注释网站，介绍基因产物的功能和结构信息，它们的生物学作用和与疾病研究的相关性。这种开放的数据资源有望使许多旨在解决导致复杂疾病及其特征成分的详细遗传图谱的项目受益。然而，由于数据结构的巨大复杂性和极高的维数，研究人员也面临着难以克服的困难。为了取得成功，继续研究开发新的计算武器库至关重要。通过概率和统计推理建模已经发现了许多引人注目的应用。