CAREER: Hierarchical Probabilistic Models for Networks with Rich Data in Scientific Domains

该项目将创建先进的算法，用于自动提取和评估现实世界网络的分层组织。网络是复杂社会和生物系统的普遍特征，包括物种或遗传相互作用网络、在线社交或离线友谊网络等等。现在，对网络的兴趣几乎涵盖了每一个科学学科。然而，越来越多地回答关于复杂网络的关键科学问题需要新的算法，这些算法可以从丰富的网络数据中自动提取其潜在的组织模式，并将这些模式与科学假设联系起来。 该项目创建了计算方法，例如，可以递归地将网络的核心与其外围分开，区分网络中的局部和全局连接模式，并将这些模式与有关顶点属性和连接强度和类型的辅助信息联系起来。特别是，该项目将使用最先进的概率生成模型来：（1）创建先进的算法，以科学家易于解释的方式自动提取连通性中层次模式的一般形式;（2）创建用于区分替代结构模式和检测随机模式的评估方法;（3）在应用于回答广泛的科学问题和数据类型时，描述它们的局限性;（4）与领域专家合作，应用这些算法来回答有关真实的网络的结构和功能的特定问题，例如，生物物种之间的相互作用，人类致病物种中的基因重组，以及人类之间的社会相互作用。该项目的成果将产生对各种社会和生物学重要领域的新见解。除了创建在科学领域具有广泛应用的先进算法外，该项目还将在网络科学的跨学科领域内培养新的研究生和本科生尖端研究技术，开发和传播新的网络科学教材，传播算法的公共领域软件实现，并与物理学、生物学、公共卫生和统计学的科学家合作解决真实的科学问题。

This project will create advanced algorithms for automatically extracting and evaluating the hierarchical organization of real-world networks. Networks are a ubiquitous feature of complex social and biological systems, including species or genetic interaction networks, online social or offline friendship networks, and more. Interest in networks now spans nearly every scientific discipline. Increasingly, however, answering key scientific questions about complex networks requires new algorithms that can automatically extract their underlying organizational patterns from rich network data, and connect these patterns with scientific hypotheses. This project creates computational methods that can, for example, recursively separate a network's core from its periphery, distinguish local and global connectivity patterns within a network, and relate these patterns to auxiliary information about vertex attributes and connection strengths and types.In particular, this project will use state-of-the-art probabilistic generative models to: (1) Create advanced algorithms that automatically extract general forms of hierarchical patterns in connectivity, in a manner that is readily interpretable by scientists; (2) Create evaluation methods for distinguishing between alternative structural patterns and for detecting random patterns; (3) Characterize their limitations when applied to answer broad classes of scientific questions and data types, and; (4) Apply these algorithms, in collaboration with domain experts, to answer specific questions about the structure and function of real networks, for example, interactions among biological species, gene recombinations in human pathogenic species, and social interactions among humans. The results of this project will generate new insights into a wide variety of socially and biologically important domains.In addition to creating advanced algorithms with broad applications across science, this project will train new graduate and undergraduate students in cutting-edge research techniques within the interdisciplinary field of network science, develop and disseminate new educational material on network science, disseminate public-domain software implementations of algorithms, and address real scientific questions in collaboration with scientists from physics, biology, public health, and statistics.