ATD:Bayesian data mining approaches for Biological threat detection

基于DNA的方法的最新发展是用于检测和表征生物制剂的可靠工具。以这种方式检测病原体是具有挑战性的，因为几乎没有遗传差异可以区分病原体与密切相关的非致病性生物体。主要研究人员提出了结合先验生物信息以及来自不同生物平台的数据的贝叶斯检测方法。基因表达微阵列数据以及大规模并行签名测序（MPSS）将通过一种新的数据融合方法相结合，以对未知数进行适当的推断。将开发基因网络模型来识别基因之间的依赖性和相互作用。主要研究者将开发分层贝叶斯模型，其中来自不同来源的数据将通过分层结构不同阶段的条件模型相互关联。他们将考虑非参数模型，这将创建一个自动聚类的基因。将局部高斯模型与Dirichlet过程先验相结合，提出了一种新的贝叶斯图聚类模型。 由于问题的复杂性，未知参数的联合后验分布将不会显式可用，因此基于马尔可夫链蒙特卡罗（MCMC）的计算方法将用于从后验分布中抽取样本。“恐怖分子很可能在未来五年内在世界某个地方使用大规模杀伤性武器，他们更有可能使用生物武器而不是核武器-结果可能是毁灭性的，”国会召集的蓝丝带小组主席在2008年12月2日告诉媒体。生物攻击比核攻击更有可能，因为它更容易实施。从历史上看，致病微生物对人类造成了损失，有时是毁灭性的。这些致病病原体可被用作生物武器。减少那里的生物制剂的科学运动之一将是发展适当检测和鉴定可用作武器的生物制剂的方法。拟议活动的知识价值在于，它将为利用基因组数据检测致命病原体提供一般和一致的框架。 通过该项目开发的有效工具和模型将用于减少这些致命病原体产生的生物威胁。这里提出的方法不仅适用于本提案中描述的场景，而且适用于各种各样的基础科学和生物医学问题的基因组数据。了解调控网络和基因相互作用将对针对细胞异常的分子治疗方法的发展产生重大影响。

Recent developments of DNA-based methods are reliable tools for detecting and characterizing biological agents. Pathogen detection in this way is challenging, because there are few genetic differences that distinguish a pathogen from a closely related nonpathogenic organism. The principal investigators propose Bayesian detection methods combining prior biological information as well as data from different biological platforms. Gene expression microarray data as well as massively parallel signature sequencing (MPSS) will be combined by a novel data fusion method to perform proper inference about the unknowns. Gene networks models will be developed to identify the dependence and interactions among the genes. The principal investigators will develop hierarchical Bayesian models where the data from different sources will be related to each other by conditional models at different stages of the hierarchy. They will consider nonparametric models which wil create an automatic clustering of the genes. Novel Bayesian graph clustering model will be developed by combining local Gaussian models and the Dirichlet process prior. Due to complexity of the problems, the joint posterior distribution of the unknown parameters will not be explicitly available hence Markov Chain Monte Carlo (MCMC) based computation methods will be used to draw samples from the posterior distribution."Terrorists are likely to use a weapon of mass destruction somewhere in the world in the next five years and they are more likely to use a biological weapon than a nuclear one -- and the results could be devastating," the chairman of the a blue-ribbon panel assembled by Congress told to media on 2nd December, 2008. Biological attack is more likely than a nuclear one because it would be easier to carry out. Historically disease-causing microbes have taken their toll on human populations, sometimes in devastating numbers. These disease causing pathogens can be utilized as biological weapons. One of the scientific movements to reduce the biological thereat will be the development of methods for proper detection and characterization of those biological agents that can be used as weapons. The intellectual merit of the proposed activity is that it will provide general and consistent frameworks for deadly pathogen detection using genomic data. The efficient tools and models which will be developed through this project will be utilized to reduce the biological threat generated from these deadly pathogens. The methods proposed here is not only applicable to the scenarios described in this proposal, but also to a wide variety of basic science and biomedical problems with genomic data. Understanding regulatory networks and gene interactions will have significant impact on the development of molecular therapeutic approaches targeted against cellular abnormalities.