Construction And Characterization Of Hypergraph-based Drug-target Hypernetwork Models

Target refers to the biological macromolecules that have pharmacological effects between drugs and human body.Its combination with drugs can play a role in the treatment of diseases.With the development of network pharmacology,scholars usually use bipartite graph to describe the relationship between drugs and targets,in which one kind of nodes are drugs and the other kind of nodes are targets.Considering the heterogeneity and diversity of nodes in complex networks,people can not effectively analyze the topological characteristics of the network,such as connectivity,cluster coefficient and degree distribution,and because the hyperedge in the hypergraph can include any number of nodes,it can more accurately describe various complex multi-dimensional relationships in the real network system.Based on hypergraph theory,this paper constructs the model,topology analysis and module mining of drug-target hypernetwork based on Hypergraph through matrix analysis and hypergraph clustering.The specific research contents are as follows:(1)Constructing a drug-target hypernetwork evolution model.Based on the realistic drug target data,this paper constructs a drug target hypernetwork evolution model based on the hypergraph structure based on the growth law of the number of targets each year,and analyzes the evolution law of target growth by mean field theory,and gets that the distribution of target hyperness is in accordance with the power law distribution,and its power index ?=1+L/M(L/M is the target growth rate),and the power index is related to the growth rate of the number of targets,and the larger it is,the faster the growth rate of the number of targets.Further,through the empirical analysis of drug-target data in Drug Bank database,the results show that the empirical and theoretical analysis and simulation results can be in good agreement.(2)Drug target hypernetwork characterization.In this paper,we collect real drug-target datasets from Drug Bank database for the construction and characterization of empirical hypernetworks.The characteristic labels of drug and target are abstracted as nodes and hyperedges respectively,and drug-target hypernetwork with drug as node and target as hyperedge,and target-drug hypernetwork with target as node and drug as hyperedge are constructed.Combining different topological indicators to quantitatively analyze the topological characteristics of the two types of hypernetworks,it was found that both types of drug-target hypernetworks have obvious scale-free features,drugs tend to connect hub target proteins,and drugs with similar functions have relatively high clustering coefficients.(3)Hypergraph clustering based on drug target modularity mining.Considering that Louvain,a clustering algorithm based on modularity maximization,performs better in terms of efficiency and effectiveness on ordinary graphs,this paper adopts IRMM(iteratively reweighted modularity maximization),a hypergraph clustering algorithm based on hypergraph modularity maximization,for module mining,and the two-section graph based drug-target network mining.The results are compared in terms of difference and functional enrichment,and the experimental results and visualization analysis show that the IRMM method is more effective for modularity mining of drug-target hypernetworks than the Louvain algorithm for drug-target bipartite graphs.