Study On Gene Regulation Modeling Algorithm Based On Non-homogeneous Dynamic Bayesian Networks

With the continuous development of microarray technology,gene expression data has become an important data source for gene regulatory network research.The use of gene expression data to construct gene regulatory networks is of great significance.For example,in medical research,gene regulatory networks can be used to identify new drug targets and design more effective therapeutic solutions.In the field of bioengineering,gene regulatory networks can also be used to design new biosynthetic routes to produce high value-added compounds.As gene expression data are often non-homogeneous,many scholars have combined multivariate processes with dynamic Bayesian networks to form non-homogeneous dynamic Bayesian network(NH-DBN),which have become the main model for reconstructing gene regulatory networks from post-genomic data.The multiple changepoint processes partitions gene expression data into disjoint segments based on the non-homogeneous of the gene regulatory network.The current commonly used sequential coupling models use similar regression parameters for each segment,and the gene regulatory networks constructed usually do not achieve the required modeling accuracy.The more advanced non-homogeneous dynamic Bayesian networks with edge-wise sequentially coupled parameters(EWC NH-DBN)model takes a different approach to the coupled and uncoupled cases.For the coupling case,in the EWC NH-DBN model,the coupling parameters are solved based on all previously coupled segments,which will lead to a decrease in the accuracy of the network reconstruction of the model.In particular,the decrease in network reconstruction accuracy becomes more pronounced as the number of segments increases.Because regulatory interactions and signal transduction processes in the cell are usually adaptive,or the responses made by cells to changes in the external environment are gradually changing.Therefore,based on the EWC NH-DBN model,this paper proposes non-homogeneous dynamic Bayesian networks with edge-wise segment-specific coupled parameters(EWSC NH-DBN)model.The model optimizes the posterior distribution of coupling parameters so that the coupling parameters of the current segment are only relevant to the previous coupling segment,which is more consistent with the regulatory interaction between cells and the signal transduction process.In addition,to avoid model over-flexibility,a two-level hyperparameter was added to the coupling parameters to enable segment-to-segment information exchange.Finally,experimental results on real gene expression data show that the EWSC NH-DBN model has better network reconstruction accuracy and stability than other classical counterparts.In the sequential coupling model.Multiple change points divide the data into mutually disjoint segments,and the parameters associated between segments are independent a priori,an assumption of a priori independence that is unrealistic in many real-world applications.The global coupling model introduces the sharing of system information between segment-specific interaction parameters.In the global coupling model,each gene node has its own independent noise variance parameter and signal-to-noise ratio parameter,but the second hyperparameter of the noise variance parameter and signal-to-noise ratio parameter is solved based on all gene nodes,and information not related to the current node will be introduced into the global information sharing,thus making the network reconstruction accuracy is reduced.Therefore,this paper proposes a parent nodes specific global coupling model on the basis of the global coupling model.In gene regulation networks,the number of parent nodes of a gene node is less than the number of all gene nodes due to the ‘fan-in constraint’,so there must be some nodes that are not the parent nodes of that node.The new model selects out the parent node of the current node during the RJMCMC iteration.The second hyperparameters of the noise variance parameter and the signal-to-noise ratio parameter are solved based on the parent node,and information about nodes that are not related to it will not be included in the global coupled model.Therefore,in this paper,based on the Globally coupling NH-DBN model,non-homogeneous dynamic Bayesian networks with noise variance and signal-to-noise parameters specific to the parent node(PS Global coupling NH-DBN)model is proposed,in which the noise variance parameter and the signal-to-noise ratio parameter are parent nodes-specific.Finally,the reconstruction accuracy of the gene regulatory network was improved,as well as the stability of the model.Finally,experimental results on real gene expression data show that the PS Global coupling NH-DBN has better network reconstruction accuracy than other classical counterparts.Figure [19] table [10] reference [74]...