Data Analysis Of Expression With Gene Microarray And Investigation For Gene Regulatory Networks

The gene microarray data is able to reveal the gene activity in various conditions and the interactions between different genes,therefore,it has a very broad application prospects for analyzing gene microarray data.Recently,the amount of gene microarray data is growing faster than the rate at which it can be analyzed and more effective techniques and methods are needed to analysis these data.In this dissertation,we introduce the key techniques and research progress of analysis of gene microarray data,and point out that many existed methods have some weakness including low capability of dealing with redundant and noisy data,unexplained mining results and limited accuracy of inferred regulatory gene networks.In light of these problems,we investigate three new gene microarray data mining methods,including feature selection and dimension reduction methods for gene microarray data,and the methods for the reconstruction of gene regulatory networks,which provide solutions for establishing classification model and identifying disease genes,and provide an effective way for biological process analysis,identifying the potential drug targets and predicting the gene function The main work and innovation of this dissertation are as follows:1.Proposing a centroid-based feature selection method.Theoretical analysis indicates that the global optimal solution of the proposed method can be reached with a non-zero initial point.In the proposed method,a kernel-based approach is used to estimate the class centroid to define the class separability criterion,based on which the objective function is formulated;2.Proposing a two-stage l1-regularized local dimension reduction method,which firstly uses a feature selection method to remove the redundant and irrelevant features,and then implements feature extraction on the selected features.The proposed method not only improves the classification accuracy,but also can generate biologically interpretable results;3.Proposing a PLS-based method for the reconstruction of gene regulatory networks(GRNs).The proposed method decomposes the GRNs inference problem into multiple feature selection subproblems,each of which is solved by a PLS-based feature selection method.Then,a statistical technique is used to refine the predictions of the inference network.Finally,Based on the technologies and methods proposed in this dissertation,a Gene Microarray Data Analysis System is designed and implemented.The system contains three main functional modules,including Feature Selection,Dimension Reduction and Gene Regulatory Network Inference,which reflect the results of our work and the practicality of this research.