Gene-disease Association Mining Based On Multi-layer Biomolecular Network

Genetic diseases are caused by the genes interaction and gene-environment interaction,and hence discovering disease gene is one of the main tasks of biomedicine.With the rapid development of high-throughput sequencing technology,the field of bioinformatics has accumulated large scale histological data,which provides new opportunities and challenges for mining the association between genetic diseases and genes.How to effectively use multiple histological data to study novel disease gene prediction methods has gradually become a research hotspot.Therefore,this paper carries out the following work:(1)We empirically analyze metabolic network and protein interaction from new perspectives.The differences in the roles of different types of genes in biomolecular networks are considered.The genes are divided into three groups including disease genes,essential genes and other genes.We analyze the topological properties and associations of the three types of genes in the metabolic network and protein interaction.Empirical results demonstrate that,compared with other genes,disease genes are topologically more important and less associate with essential genes.(2)We construct a two-layer biomolecular network with a protein interaction layer and a metabolic layer.We design a novel disease gene sequencing method for the bilayer biomolecular network.Based on the classical hypothesis that the neighbors of disease genes trend to cause similar disease,positive flow is assigned to disease genes,while negative flow is assigned to essential genes based on the empirical analysis that disease genes are less associated with essential genes.Based on two-layer biomolecular networks,synchronous network propagation method and asynchronous network propagation method are proposed respectively.The effectiveness and potential of the new method are verified by 110 genetic diseases,and the performance of the new method is particularly good in monogenic genetic diseases.(3)In order to make full use of the topological features of multilayer biomolecular networks,a disease gene method based on graph convolutional neural is proposed.The network representation learning algorithm Node2 Vec is adapted to extract low-dimensional factors of nodes based on the typical assumption that the neighbors of disease genes are likely to cause diseases and the disease genes are less associated with essential genes.In addition,the attention mechanism is introduced into the graph convolutional neural network to effectively capture the topological correlation of different nodes.The experimental results show that the prediction performance of the proposed method is significantly better than traditional methods,such as Dis Mult,and the AUC and F1-score can achieve 0.872 and 0.843.