III: Small: Topology-based approaches to integrated analysis of transcriptomic, protein interactomic and phenotypic data

高通量技术现在允许同时测量细胞中数万个分子的活动和相互作用，为生物学中的系统级科学探索打开了新的大门。先进的计算方法正在开发中，以分析海量数据，提取代表知识的模式，并构建预测模型，有望确定生物体的特征，如癌症结果或植物生长表型。为了帮助实现这些目标，该项目旨在开发高效和有效的计算算法和工具，以整合异质和噪声的高通量数据，并以将基因视为细胞中相互关联的组件而不是独立组件的方式进行分析。生物网络是描述细胞内分子间相互作用的数学模型，对复杂生物系统的建模和理解至关重要。然而，生物网络的庞大和复杂，以及噪声和不完整的数据，给基于网络的数据分析带来了严峻的挑战。为了应对这些挑战，一个实用和直观的策略是在功能通路的水平上分析/利用这种网络，即参与相似生物过程的基因/蛋白质，这将显著降低生物网络的复杂性，并提高对复杂表型的理解。由于目前对大多数物种的功能通路的了解相当有限，本项目将开发一套算法和软件工具，用于完全自动化地发现密集的子网络作为候选功能模块，并开发面向功能模块的算法，用于分析/利用生物网络的几个实际应用。首先，该项目将开发算法来提高网络质量，并利用网络拓扑中嵌入的信息来发现网络模块。对于网络化(例如蛋白质-蛋白质相互作用)数据，利用拓扑结构来提高边的可靠性，并使用一种新的基于图上随机游走的拓扑相似性度量来实现模块发现。对于非联网的(例如，转录的)数据，利用全局网络拓扑来构建能够在没有任何用户指定的参数的情况下实现全自动模块发现的“最佳”网络。其次，这项研究将开发计算方法来系统地研究网络拓扑和生物功能之间的关系，这有望促进当前对生物网络组织原理的理解，并有助于在疾病研究中确定基因的优先顺序。最后，本项目提出了一种新的基于Steiner树的算法，用于识别与癌症表型相关的潜在致病基因，以及一种新的相似性度量，用于基于路径/子网络级别的基因表达模式比较患者，该算法可以很容易地与现有的基于网络的聚类/分类算法相结合，用于基于网络的癌症预后预测。该项目的最终产出将包括用于综合数据分析的生物信息学工具和从不同输入数据集中发现的生物知识数据库。这些工具和资源将在网上免费提供，可供对生物信息学算法开发或应用感兴趣的广泛研究人员使用。这些工具和资源将用于研究合作者最感兴趣的几个生物过程，合作者已承诺验证一些计算预测。这些包括识别新的植物激素反应基因，预测和表征DNA损伤反应基因，以及通过整合蛋白质-蛋白质相互作用和转录数据预测乳腺癌患者的转移潜力。该项目还将通过开发新的网络链接预测和模块发现算法以及网络约束的聚类/分类方法来促进计算的进步，预计这些方法将立即在生物科学以外的其他领域得到应用。作为这项研究的一部分开展的活动将被纳入几门课程，并将扩大德克萨斯大学圣安东尼奥分校的教育和研究机会，这是一所为少数群体服务的机构，大多数本科生来自代表性不足的少数群体，预计将增加地理和种族多样性，并鼓励少数群体参与生物信息学和计算生物学研究。

High-throughput technology now allows measuring the activities and interactions of tens of thousands of molecules in the cell simultaneously, opening new doors to systems-level scientific exploration in biology. Advanced computational methods are in development to analyze the huge amount of data to extract patterns that represent knowledge and to construct predictive models that have the promise to determine the characteristics of an organism, such as cancer outcomes or plant growth phenotypes. To help achieve these goals, this project aims at developing efficient and effective computational algorithms and tools to integrate heterogeneous and noisy high-throughput data, and to analyze them in ways that treat genes as inter-connected rather than independent components of the cell. Biological networks are mathematical models that describe the interactions among molecules in the cell and are critical to the modeling and understanding of complex biological systems. However, the large sizes and complexity of biological networks as well as the noisy and incomplete data pose critical challenges to network-based data analysis. To tackle these challenges, a practical and intuitive strategy is to analyze/utilize such networks on the level of functional pathways, i.e., genes/proteins involved in similar biological processes, which would significantly reduce the complexity of biological networks and improve the understanding of complex phenotypes. As the current knowledge of functional pathways is rather limited for most species, this project will develop a set of algorithms and software tools for fully automated discovery of dense subnetworks as candidate functional modules, and develop functional module-oriented algorithms for analyzing/utilizing biological networks for several real applications. First, this project will develop algorithms to improve network quality and network module discovery using information embedded in network topology. For networked (e.g. protein-protein interaction) data, topology is utilized to improve edge reliability, and subsequently module discovery, using a novel topological similarity measurement based on random walks on graphs. For non-networked (e.g., transcriptomic) data, global network topology is utilized to construct an "optimal" network that enables fully automated module discovery without any user-specified parameters. Second, this research will develop computational methods to systematically investigate the relationship between network topology and biological functions, which is expected to advance the current understanding of the organizing principles of biological networks, and facilitate prioritizing genes in disease studies. Finally, this project proposes a novel Steiner tree based algorithm for identifying potential causal genes associated with cancer phenotypes, and a novel similarity metric to compare patients based on pathway/subnetwork-level gene expression patterns, which can be easily combined with existing clustering/classification algorithms for network-based prediction of cancer outcomes. The final outputs of this project will include both bioinformatics tools for integrative data analysis and databases of biological knowledge discovered from different input datasets. These tools and resources will be made freely available on the web, which can be used by a broad range of researchers who are interested in bioinformatics algorithm development or applications. These tools and resources will be applied to study several biological processes of central interests to collaborators, who have committed to validate some of the computational predictions. These include identifying novel plant hormone response genes, predicting and characterizing DNA damage response genes, and predicting metastasis potentials for breast cancer patients, by integrating protein-protein interaction and transcriptomic data. This project will also contribute to the advancement of computing with the development of novel network link prediction and module discovery algorithms and network-constrained clustering/classification methods that are expected to have immediate applications in other domains besides biological sciences. The activities undertaken as part of this research will be incorporated into several courses and will expand the educational and research opportunities available at the University of Texas at San Antonio, a minority-serving institute where the majority of undergraduates are from under-represented minorities, and is expected to increase the geographic and ethnic diversity and encourage the participation of minority groups in bioinformatics and computational biology research.