Research On Multifactor Model Optimization And Inference Methods For Gene Regulatory Networks

Gene regulatory networks can help researchers to gain insight into the interactions and regulatory mechanisms among genes in biological systems,thereby revealing the essence and organizational structure of life phenomena,it is critical for predicting and diagnosing the occurrence and evolution of diseases,as well as developing new therapeutic methods and drug targets.Since the advent of high-throughput sequencing technologies,researchers have access to an increasing amount of gene expression data,which has led to the development of various models and algorithms for reconstructing gene regulatory networks to mine potential gene regulatory relationships.Various algorithms for inferring gene regulatory networks based on computer technology have been developed with the advancement of information science.Gene regulation is a complex process that is influenced by multiple factors,including the characteristics of the genes themselves and the regulatory mechanisms,such as gene decay and regulatory time delay.Despite the crucial importance of these factors for inferring gene regulatory networks,the existing nonlinear differential equation models have not systematically studied and considered these two factors,and there is a lack of a unified and effective optimization method to estimate these key parameters.Aiming at the above problems,this thesis discussed about the gene regulatory network model,constructed a nonlinear differential equation model that combines gene decay and regulatory time delay to investigate the effects of these two factors,and proposed a gene regulatory network inference algorithm,GRNMOPT,based on multi-objective optimization methods to improve the accuracy of gene regulatory network inference.The main work of this thesis is as follows:(1)Investigating the impact of gene decay and regulatory time delay on the accuracy of gene regulatory network inference.We established a nonlinear differential equation model,setting decay rate c and time delay? to simulate gene decay and regulatory time delay among genes.Specifically,we studied three scenarios during modeling,not considering c and?,single-factor(c or?)effects,and coordinated effects of the two factors,to gain a deeper understanding of the mechanisms of these key factors in gene regulatory network models.Experiments were carried out on the simulation and real data sets(DREAM4 In Silico＿Size10and Yeast)to comprehensively and accurately evaluate the impact of these key factors on gene regulatory network inference,which can provide guidance for model optimization.(2)A gene regulatory network inference algorithm called GRNMOPT based on a multi-objective optimization method was proposed.Firstly,GRNMOPT utilizes the decay rate and the time delay jointly to construct a nonlinear differential equation model based on steady-state data and time-series data.Moreover,the machine learning algorithm XGBoost is applied to learn the nonlinear function,which efficiently and accurately calculate the regulatory strength between genes.Finally,GRNMOPT uses a multi-objective optimization strategy based on the Non-dominated Sorting Genetic Algorithm II(NSGA-II)to simultaneously optimize the decay rate and time delay,and obtains the Pareto optimal set of two parameters to maximize the inference accuracy.We compared our method with four state-of-the-art algorithms,GENIE3,Bi XGBoost,Nonlinear＿ODE,and MMFGRN,on different simulated and real datasets(DREAM4,yeast,and Escherichia coli datasets).Extensive experimental results show that the proposed method exhibits encouraging performance on different scale of networks.Further,we conducted the cross-validation experiments on the DREAM4 In Silico＿Size100 and Escherichia coli datasets to verify the robustness and reliability of our proposed method.