Computational Methods for Functional Genomic Discovery from Gene Knockout Studies

描述(申请人提供)：第二阶段的总体目标是继续开发/验证新的系统生物学计算工具，用于从多扰动基因敲除实验获得的基因表达数据推断基因调控关系。美国国立卫生研究院的基因敲除小鼠计划(KOMP)是一项旨在产生小鼠胚胎干细胞(ES)的公共资源的倡议，该资源包含小鼠基因组中每个基因的零突变--这对于破译小鼠乃至人类生物系统的复杂性非常重要。预计生物医学研究的所有领域都将出现新一代生物系统视角的多重扰动/KO研究。破译基因调控反应(基因-表型信号级联)的新计算工具将大大有助于我们对许多疾病的分子靶标和机制的理解。今天，研究人员需要新的工具来处理和破译从多扰动研究中产生的海量基因/蛋白质表达数据。Seralogix的第二阶段工作重点是改进和创建用于学习更大规模(生物系统级别)基因调控网络的新功能，并将此网络学习功能整合到我们现有的生物系统分析框架(BAF)中。我们的BAF由一套集成的数学分析和建模工具以及数据库组成。BAF核心工具基于动态贝叶斯网络(DBN)。DBN允许我们系统地将先验知识与经验的时间历程表达数据相结合，用于建模、模式识别以及最终的生物系统遗传网络学习，如本文所建议的。我们的算法创新，在第一阶段被证明是可行的，是将生物学先验知识和多扰动数据与我们的DBN结合起来，以实现遗传网络学习方法。这种方法是基于成熟的贝叶斯统计方法，我们在抽样方案中采用了生物先验知识来克服从稀疏和噪声基因表达数据中进行结构学习的固有困难。在第一阶段，我们展示了先验知识，结合贝叶斯网络学习方法和多扰动/KO实验数据，导致了可靠的基因调控关系识别。我们相信，这种方法可以扩大规模，从而产生更稳健的数学/功能系统级模型。此外，我们相信，将遗传网络学习整合到Seralogix的BAF中将为识别新的基因调控关系和对疾病过程的洞察提供重要的新工具，并对Seralogix具有巨大的商业潜力。我们将与德克萨斯基因组医学研究所合作，提供老鼠基因表达KO数据，他们正在研究出生缺陷的基因组原因。我们的第二阶段目标包括：1)扩展我们的方法以支持生物系统级别的网络学习；2)对我们学习的网络进行统计评估和生物验证；3)开发新的工具/技术来询问所产生的系统网络模型，以便生物学家可以提取重要的知识。 与公共卫生相关：充分阐明控制健康和疾病的分子决定因素的复杂相互作用和调节，这是现代生物学研究的最终目标之一，仅举几个例子，细胞周期、发育生物学、衰老以及复杂疾病的进行性和复发性发病机制。拥有新的计算方法(软件工具)来识别和破译受基因调控的反应(例如，信号级联反应)将大大有助于促进我们对许多高度公共卫生关切疾病的分子靶标和机制的理解。发现潜在的遗传功能和关系对于取得医学突破，特别是对于安全和有效的药物和诊断开发将是极其重要的。今天，研究人员受到来自基因敲除研究产生的海量基因/蛋白质表达数据的阻碍。通过数学建模将这些原始基因组/蛋白质组数据转化为可操作知识的计算工具将有助于指导和加快研究人员对遗传疾病的调查，并确定干预和治疗的目标。

DESCRIPTION (provided by applicant): The overall Phase II objective is to continue the development/validation of new system biology computational tools for inferencing gene regulatory relationships from gene expression data obtained from multi-perturbation gene knockout experiments. NIH's Knockout Mouse Project (KOMP) is an initiative to generate a public resource of mouse embryonic stem (ES) cells containing a null mutation in every gene in the mouse genome - important for deciphering the complexity of biological systems of mice and ultimately man. It is anticipated that a new generation of multi-perturbation/KO studies with a biological system perspective will emerge in all areas of biomedical research. New computational tools for deciphering genetically regulated responses (genotype-to- phenotype signaling cascades) will significantly aid in advancing our understanding of the molecular targets and mechanisms of many diseases. Today, researchers need new tools to deal with and decipher the tremendous volumes of gene/protein expression data generated from multi-perturbation investigations. Seralogix's Phase II efforts focus on improving and creating new functionality for learning larger scale (biological system level) gene regulatory networks and integrating this network learning functionality into our existing Biosystem Analysis Framework (BAF). Our BAF is comprised of a suite of integrated mathematical analysis and modeling tools and databases. The BAF core tools are based on Dynamic Bayesian Networks (DBNs). DBNs allow us to systematically integrate prior knowledge with empirical time-course expression data for modeling, pattern recognition and eventually biological system genetic network learning as proposed herein. Our algorithmic innovation, proven feasible in Phase I, is the incorporation of biological prior knowledge and multi-perturbation data with our DBNs for enabling a genetic network learning approach. This approach is based on well established Bayesian statistical methods that we adopt in a sampling scheme enhanced with biological prior knowledge to overcome the intrinsic difficulty of structure learning from sparse and noisy gene expression data. We show in Phase I that prior-knowledge, coupled with Bayesian network learning methods and multi-perturbation/KO experimental data, resulted in reliable gene regulatory relationship identification. We believe this approach can be scaled up, leading to a more robust mathematical/functional system level model. Further, we believe that integrating genetic network learning into Seralogix's BAF will provide an important new tool for identifying novel gene regulatory relations and insights into disease processes and have significant commercial potential for Seralogix. We will be collaborating with the Texas Institute of Genomic Medicine as a provider of mouse gene expression KO data who are studying the genomic causes of birth defects. Our Phase II aims include: 1) scaling our approach to support biological system level network learning; 2) statistical assessment and biological validation of our learned networks; 3) developing new tools/techniques to interrogate the resulting system network models so biologist can extract important knowledge. PUBLIC HEALTH RELEVANCE: It is one of the ultimate goals for modern biological research to fully elucidate the intricate interplays and the regulations of the molecular determinants that control health and disease, to name a few, cell cycling, developmental biology, aging, and the progressive and recurrent pathogenesis of complex diseases. Having new computational methods (software tools) for identifying and deciphering genetically regulated response (e.g. signaling cascades) will significantly aid in advancing our understanding of the molecular targets and mechanisms of many diseases of high public health concern. The discovery of underlying genetic function and relationships will be extremely important for making medical breakthroughs, especially for the safe and effective development of drugs and diagnostics. Today, researchers are hindered by the tremendous volumes of gene/protein expression data generated from knockout investigations. Computational tools that transform these volumes of raw genomic/proteomic data to actionable knowledge via mathematical modeling will help guide and accelerate researchers' investigations of genetic disorder and identifying targets of intervention and treatment.