Centrality Rankings In Multilayer Networks

Centrality ranking of nodes or layers in multilayer networks is an essential research topic,it quantifies the importance of influential nodes or layers in complex networks and is widely used to characterize the structure and function of multilayer systems.In the era of big data,the centrality rankings are required in a plethora of applications,such as information retrieval and spread,disease and link prediction,marketing and advertising strategies.The existing centrality rankings in multilayer networks cannot efficiently identify important nodes and lack of considering the influence of layer importance on the ranking of nodes.Therefore,this thesis conducts research and develops novel centrality rankings,respectively,in multiplex networks with/without layer couplings,interconnected multilayer networks and multilayer heterogeneous networks,aiming at accurately and efficiently identifying the essential nodes or layers.The main contributions and innovative achievements are summarized as follows.(1)Centrality ranking of nodes in multiplex network without layer couplings is proposed.Current centrality rankings of nodes in multiplex network without layer couplings focus only on either diffusion processes based on random walks or on the topological structure of networks.Considering these two aspects,this thesis first proposes a general expression of topologically random walks;then,in terms of such random walk,proposes a topologically biased multiplex PageRank to characterize the centrality ranking of nodes in multiplex networks without layer couplings.In particular,according to the nature of biases and the interaction of nodes between different layers,the additive,multiplicative and combined biased multiplex PageRank are distinguished,revealing how each case reflects the impact of ranking of nodes in one layer on those of their replicas in another layer and captures the extent to which the walkers preferentially visit either hubs or poorly connected nodes by tuning the bias parameters.Experimental results show the proposed approach can efficiently capture the significantly top-ranked nodes by properly tuning the biases in the walks.(2)Co-ranking for nodes and layers in multiplex network with layer couplings is presented.Current centrality rankings in multiplex network only give the ranking of nodes and almost never concern the influence of layer importance on the ranking of nodes.Moreover,some centrality rankings are not applicable to multiplex network with large layers.This thesis proposes a co-ranking algorithm for the ranking of nodes and layers in large multiplex networks with layer couplings.Co-ranking considers the full multiplex network structure and capitalizes on the dual nature of nodes and layers,aiming at assigning more centrality to nodes receiving links from already central nodes and from highly influential layers.The algorithm can apply to directed/undirected,weighted/ unweighted multiplex networks.(3)Centrality ranking based on a tensor framework in interconnected multilayer networks is proposed.Not all nodes in real-world networks are shared by all layers,therefore,identifying the essential nodes in interconnected multilayer networks is crucial for understanding their topology and dynamic processes.To this end,this thesis proposes a centrality approach based on tensorial framework to characterize the ranking of nodes in interconnected multilayer networks.This approach can quantify the relationship between the node centralities and layer influences,and flexibly integrate prior knowledge of the interaction among layers to obtain the tailored centrality of nodes,accounting for how the centrality of nodes propagates among distinct layers.(4)Centrality ranking of nodes in multilayer heterogeneous biological networks is presented.Current centrality rankings in multilayer networks are based on multiple types of links,but each layer may be a heterogeneous bipartite graph network,for example,a multilayer heterogeneous network comprises of a multilayer gene network,a phenotype network and a gene-phenotype network,where the genes of each layer are connected to their related phenotypes according to the known bipartite associations,the centrality based on such network type is an unresolved issue.This thesis proposes a topologically biased random walk with restart(BRWR)algorithm applicable to such network for the identification of disease genes and candidate gene ranking.Experimental results show that the BRWR algorithm applied to multilayer heterogeneous networks can reliably obtain the ranking of the candidate disease genes,predict disease genes implicated in the undiagnosed neonatal progeroid syndrome and explore network representation of the SHORT syndrome and its associated PIK3R1 gene.