Brain knowledge graph analysis based on complex network theory

Abstract

Domain knowledge about the brain is embedded in the literature over the whole scientific history. Researchers find there are intricate relationships among different cognitive functions, brain regions, brain diseases, neurons, protein, gene, neurotransmitters, etc. In order to integrate, synthesize, and analyze what we have known about the brain, the brain knowledge graph is constructed and released as part of the Linked Brain Data (LBD) project, to reveal the existing and potential relationships of brain related entities. However, there are some incorrect and missing relationships in the extracted relations, and researchers also cannot find the key topics overwhelmed in the massive relations. Some researchers analyze the properties of vertices based on the network topology, but they cannot verify and infer the potential relations. In order to address the above problems, we propose a framework which consists of 3 parts. Firstly, based on complex network theory, we adopt the embeddedness to verify the relations and infer the potential links. Secondly, we use the network topology of existing knowledge to build the self-relations graph. Finally, the structural holes theory from sociology is adopted to discover the key and core vertices in the whole brain knowledge graph and we recommend those topics to users. Compared with logic inference methods, our methods are lightweight and capable of processing largescale knowledge efficiently. We test the results about relation verification and inference, and the result demonstrates the feasibility of our method.