面向知识发现的网络对齐方法及其在生物网络中的应用

Large-scale network data with complex internal structure are emerging rapidly in many areas, such as bioinformatics and social network. However, rich knowledge still lies behind these data and remains undiscovered. Therefore, there is an urgent need to effectively support the knowledge discovery among such large-scale and complex network data and well address the significant challenge. Network alignment across multi-networks provides a new approach for knowledge discovery in network data to exploit the knowledge that is not feasible from mining in a single network. However the research of large-scale network alignment is still in its infancy, which needs more efforts in further development and improvement. This project aims at developing and improving network alignment approaches to support better knowledge discovery in various applications. First, we study the establishment of similarity measure model for large-scale multi-networks. Then we develop effective algorithms of network alignment and supportive techniques of network alignment parallelization. Algorithms developed in this project will be combined with multi-network cooperative partition strategy to conduct highly efficient and effective parallel computations. Finally, proposed approaches and techniques will be applied and validated in several applications of different areas in discovering valuable knowledge. The research results from this project will provide powerful supports to the applications of network alignment for knowledge discovery and largely promote the development of the bioinformatics areas.

随着生物信息、社交网络等领域涌现规模庞大且内部结构复杂的网络数据，如何有效的挖掘这些数据背后蕴藏的丰富知识成了知识发现技术面临的重大挑战。网络对齐通过建立多个网络之间的信息流，为跨网络知识发现提供了一种新途径，但是针对大规模网络的对齐研究正处于起步阶段，需要进一步发展和完善。本项目将研究面向知识发现的网络对齐方法。首先研究设计多网络相似性度量方法，建立网络对齐的理论模型；然后重点研究网络对齐算法，针对大规模网络的对齐设计高效、准确的算法，在算法设计中考虑复杂网络数据的异质性，并结合多网络协同划分方法实现网络对齐的并行化；最后结合生物网络数据研究网络对齐的应用，为生物信息等领域的知识发现提供技术支持。项目研究成果将为网络对齐技术在知识发现中的应用提供科学方法，继而推动相关领域知识发现的发展。

大规模图数据中蕴含着丰富的知识，如何有效挖掘多个图之中的知识是一个重要挑战。网络对齐将不同图中的节点或子图进行映射，建立多个网络之间的信息流，从而提供了一种在多个图中进行知识发现的新途径。本项目研究了面向知识发现的网络对齐方法及其应用，主要成果如下：（1）针对网络对齐任务提出了网络相似性度量模型；（2）提出了面向知识发现的网络对齐算法，在不同物种的蛋白质相互作用网络和跨语言实体对齐中进行了验证；（3）针对大规模图网络提出了图神经网络结构搜索方法；（4）应用相关理论和方法，设计了针对生物网络等不同类型图数据的应用模型。本项目的研究为大规模图数据挖掘提供了理论和技术支持，推动了图数据挖掘技术在生物医药等领域的应用发展。