Dynamic Bayesian Network Modeling Based on Structure Prediction for Gene Regulatory Network

Abstract

Gene regulatory network can intuitively reflect the interaction between genes, and an in-depth study of these relationships plays a significant role in the treatment and prevention of clinical diseases. Therefore, correct reconstruction of gene regulatory network has become the first critical step in the study of disease treatment and prevention at the genetic level. Among the methods for gene regulatory network reconstruction, the Bayesian network model has been widely concerned because of its advantages of expressing both the regulatory relationship and the degree of strength between genes. Nevertheless, the complexity of the Bayesian network model in structure learning is extremely high, making the efficiency of the reconstruction network is low and the scale is limited. Therefore, this paper proposed a dynamic Bayesian network modeling based on structure prediction (DBN-SP). The method combines the correlation model with the dynamic Bayesian network model. On the premise of making full use of the search strategy of dynamic Bayesian network model structure learning, the candidate parent node set is selected based on the structure prediction firstly. Based on this, some redundant information can be removed and the search space can be reduced in the DBN structure learning to improves the efficiency of the network reconstruction. After the network is reconstructed, structure optimization by using the conditional mutual information method can further remove redundant edges and make the network more accurate. The experimental results show that DBN-SP greatly improves the efficiency and scale of the gene regulatory network reconstruction, and the accuracy and other indexes are also improved. DBN-SP is freely accessible at https://github.com/quluxuan/DBN-SP.git .