Design And Implementation Of Algorithms For Community Detection In Biological Network Based On Graph Embedding

In recent years,the application of network biology in the field of biomedicine has gradually increased.In clinical and ecological research,network biology is an effective means to identify the interdependence between individual organisms and explore the functional communities of organisms.However,at present,in order to achieve the identification of the network structure in the complex biological network and the precise detection of biological communities with biological significance,there is an urgent need for more excellent community discovery algorithms.As a type of emerging technology,graph embedding algorithms have shown excellent performance in the exploration of network characteristics.Although the algorithms have been successfully applied to various biomedical researches,the exploration of their application potential in the detection of biological communities is insufficient.Whether the graph embedding algorithms are suitable for the community detection in biological networks,how to set the parameters and the real performance still need to be further discussed.The study takes research on community detection in biological networks based on the idea of combining graph embedding with clustering.Specifically,the article applies a variety of classical graph embedding algorithms based on random walk,matrix factorization and neural networks combining with different clustering algorithms to discover biological communities.The downstream clustering algorithms are improved to obtain biological communities of suitable size based on the sizes of biological functional pathways.In the testing phase,the performance of the community detection algorithms are scored with new standard,too.Based on six different types of network data sets and multiple disease-related genome-wide association analysis data,the performance of algorithms are scored by the association significance between the communities and the diseases.By comparing the score of algorithms using graph embedding with multiple classic community detection algorithms,we objectively evaluate the performance of graph embedding algorithm in detecting communities in biological networks.The results of this study show that the algorithms for detecting bio-communities based on graph embedding have excellent performance,which outperform many types of classic network-community detection algorithms.Though the graph-embedding biocommunity detection algorithms show different performance in different biological networks,but the overall performance are relatively good.Finally,the study applies the optimized K-Means + Deep Walk algorithm,which has been shown to be suitable for community detection in co-expression network in the paper,to the COVID19 data set.Compared with the classic algorithm WGCNA,the community detection algorithm based on graph embedding gets more refined gene communities.Further interpretation of the biological function of each community shows that these communities are related to the clinical symptoms of COVID19.The function of some communities can provide novel conjectures for COVID19 research and guidance for clinical research.