Inference Of Gene Regulation Network Dynamics Model Based On Clustering And Mv-MOEA Algorithm

Inferring the gene regulatory network is one of the important ways to study biological information.It can explore the potential mechanism of the cellular system,and its accuracy directly affects the accuracy of biological genetic information.This paper mainly studies the application of clustering algorithm and multi-objective optimization algorithm in gene regulation network inference.The main work is as follows:Aiming at the problem of parameter estimation in inferred network,a binary variable representing network connection is introduced.A multi-objective optimization algorithm of mixed variable is proposed.The data fitting error and_?-norm are taken as two targets in the biological target optimization model.The mixed variable multi-objective optimization algorithm obtains the non-dominated solution of the biological target optimization model,and then selects the final inference result from the non-dominated solution through the automatic trade-off selection process.Experiments on the artificial data set show that the mixed variable multi-objective optimization algorithm can effectively infer the small gene regulatory network in low noise environment.In view of the ubiquitous dimension explosion of gene expression data in large gene regulatory networks,an optimal PSO(Partial Swarm Optimization)-k-means clustering algorithm was proposed to rapidly cluster gene expression data.Firstly,the number of clusters was determined by using contour coefficients.The value of k is studied to make the clustering result of gene expression data more reasonable.When the optimal value of k is determined,the gene expression data is clustered by PSO-k mean algorithm to achieve a reasonable segmentation of large gene regulatory networks,reduce the complexity of the inference process.In order to solve the problem of no connection and independence between the divided sub-networks,the concept of representative genes is further proposed,and each sub-network is represented by Genes are linked to each other.Screening for regulatory genes of target genes,a secondary screening method was adopted to make the selected regulatory genes more comprehensive.Firstly,the neighboring genes of the target genes were screened according to the distance from the genes in the same sub-network to the target genes,and the representative gene is composed of candidate genes,and then the parameters of the candidate genes are solved by a mixed variable multi-objective optimization algorithm,and the threshold is set to remove the genes with smaller parameters in the candidate genes.Experiments on the gene expression data set of Saccharomyces cerevisiae show that The regulatory genes obtained from the secondary screening are reasonable,and the gene expression data inferred from the S-system model is similar to the original data.In this paper,some researches and explorations on the dynamics model of gene regulatory networks have been carried out.A hybrid variable multi-objective optimization algorithm is proposed to infer small-scale gene regulatory networks,and a large-scale multi-objective optimization algorithm based on clustering and mixed variables is proposed.Gene regulation network inference method,and proved its feasibility on the real data set.