Research On Gene Regulatory Network Reconstruction Based On Dynamic Bayesian Network

With the continuous development of bioinformatics,biological gene regulatory networks,protein cascades have gradually become the focus of research,and detailed and in-depth exploration of gene regulatory networks can help humans to better grasp biological information.Therefore,inferring the structure of gene regulatory networks from data has become the focus of bioinformatics research.Currently,there exist several methods to model the network structure,and non-homogeneous dynamic Bayesian network is one of them.Hidden Markov Model Non-homogeneous Dynamic Bayesian Network(HMM-DBN),which can describe both gene regulation relationship and gene regulation direction,has become an effective method to construct gene regulation network.In this thesis,based on HMM-DBN,different information coupling methods are studied and compared to improve the accuracy of network reconstruction and design effective inference methods,and the specific work includes the following aspects:Because the traditional HMM-DBN model assumes that the parameters are completely independent,which makes the parameters in the time period need to be sampled separately,ignoring the adaptability and continuity of the process of biological adaptation to environmental changes and affecting the accuracy of network reconstruction.To address this problem,the Global Coupling Hidden Markov Model Non-homogeneous Dynamic Bayesian Network(GCHMM-DBN)with global coupling of parameters is proposed in this thesis.The GCHMM-DBN model is mainly based on the HMM-The GCHMM-DBN model is mainly based on the HMM-DBN for information interaction,thus improving the parameter independence.On the one hand,the global interaction hyperparameters vector is constructed by variance hyperparameters and signal-to-noise hyperparameters,so that the parameter information of each node is shared in all time periods.On the other hand,experiments on the information coupling methods among the three nodes demonstrate that the information sharing methods among the nodes affect the network structure inference.The experimental results on yeast dataset and synthetic RAF dataset show the improvement of network reconstruction accuracy over HMM-DBN.Since the global coupling approach forces the parameters to remain coupled for all time periods,this global coupling approach is clearly not practical when the environment or experimental conditions change significantly.Therefore,in order to consider the complex process of changing regulatory relationships between genes,it is necessary to properly maintain the information independence between gene nodes.To address the above problems,this thesis proposes Edge-wise Coupling Hidden Markov Model Non-homogeneous Dynamic Bayesian Network(EWCHMM-DBN)based on edge-coupling.EWCHMM-DBN determines from the data whether the current segment should be coupled to the previous segment,samples the regression parameters based on the coupling of the segments,and inferred the gene regulatory network by combining the regression parameters and the time segments.The experimental results on yeast dataset and synthetic RAF set show that EWCHMM-DBN can predict the network structure better than HMM-DBN.Figure [18] table [6] reference [79]...