Dynamic Network Analysis: Analyzing the Chronnectome

网络描述对象之间的互动：例如，它们可以描述人与人之间的联系，如在社交网络中；捕获生物过程中的依赖关系，如基因调控；或者代表大脑中不同结构和功能区域的相互作用和连接。使用不同的定量方法开发了许多网络分析方法。然而，这些方法中的许多仍然主要用于静态网络的分析，即在特定时间点的网络快照。然而，为了评估诸如大脑中的网络如何随时间变化，需要改进动态网络分析的方法。该项目将开发数学方法，以驱动生物学问题来分析大脑结构和功能连接的时间变化，即时间组。结构连接包括大脑中的白质连接，从而实现信息交换的底层布线。另一方面，功能性大脑连接描述了任务如何影响大脑活动，以及不同的大脑区域在任务下如何表现相似。随着时间的推移，对结构和/或功能连通性的变化进行量化可以提高对偏离正常的脑部疾病的理解。此外，这里开发的方法将普遍用于其他动态网络，从其他生物网络到社会网络等等。为了最大限度地发挥作用，这里开发的计算方法将以开源的形式提供，软件许可证允许免费的商业和非商业用途。该项目还包括针对网络科学学生的研究培训和课程。网络通常被描述为图，节点描述系统中的实体，边描述节点关系。网络的范围从非结构化和快速变化的社会网络到结构化的缓慢变化的网络，这些网络捕获了结构性的大脑连接。网络科学寻求开发从交互模式中挖掘信息的方法，例如，在社区中提取紧密耦合的节点。随时间变化的网络数据越来越多；然而，由于大多数方法都集中在静态数据上，因此仍然缺乏足够的分析方法，而时间依赖网络的方法的持续发展直到最近才变得更加普遍。该项目的目标是推进一般的时间依赖网络分析，以脑时间组分析为导向驱动问题，推动技术发展。目前的时间依赖网络分析方法缺乏时间组学分析所需的几个关键特性：例如，(i)纵向网络数据分析的能力，其中网络可用于多个时间点的多个受试者，（ii）包含特定领域先验信息的能力（例如连接模式的先验知识），以及（iii）处理非同质受试者群体和时间不连续性的能力。为了解决这些缺点，该项目将基于(i)网络分析随机块模型方法的扩展和（ii）网络值数据的回归模型开发定制的网络分析方法。

Networks describe interactions between objects: they can for example describe connections between people, as in social networks; capture dependencies in biological processes such as gene regulation; or represent the interplay and connectivity of different structural and functional areas in the brain. Numerous methods for network analysis have been developed using different quantitative approaches. However, many of these methods are still predominantly available for the analysis of static networks, i.e., snapshots of a network at a particular point in time. However, to assess how networks such as those in the brain change over time, improved methods for dynamic network analysis are required. This project will develop mathematical approaches motivated by the driving biological problem of analyzing temporal change of structural and functional brain connectivity, the chronnectome. Structural connectivity includes white matter connections in the brain and hence the underlying cabling enabling information exchange. Functional brain connectivity on the other hand describes how tasks influence brain activity and how different brain areas behave similarly under tasks. Quantifying changes in structural and/or functional connectivity over time can improve understanding of brain diseases, which deviate from normality. Moreover, the approaches developed here will have general use for other dynamic networks, from other biological networks to social networks and beyond. To maximize impact, the computational methods developed here will be made available in open-source form, with a software license permitting free commercial and non-commercial use. The project also includes research training and courses for students in network science.Networks are commonly described as graphs, with nodes describing entities in a system and edges node relationships. Networks range from unstructured and rapidly changing social networks to structured slowly varying networks capturing structural brain connectivity. Network science seeks to develop methods to mine information from interaction patterns, for example, extracting tightly coupled nodes in communities. Time-dependent network data is increasingly available; however, sufficient analysis methods are still lacking as the majority of approaches have focused on static data, with ongoing development of methods for time-dependent networks having become more common only recently. The goal of this project is to advance general time-dependent network analysis, with brain chronnectome analysis as the guiding driving problem to motivate the technical development. Current approaches for the analysis of time-dependent networks lack several key properties required for chronnectome analysis: e.g., (i) the ability to analyze longitudinal network data, where networks are available for multiple subjects at multiple timepoints, (ii) the ability to include domain-specific prior information (such as prior knowledge of connectivity patterns), and (iii) the ability to deal with inhomogeneous subject groups and discontinuities in time. To address these shortcomings, this project will develop customized network analysis approaches based on (i) extensions to the stochastic block model approach of network analysis, and (ii) regression models for network-valued data.