Research And Application Of Module Biomarker Search Algorithms In Biological Network

With the development of high-throughput technology,a large number of omics data have been produced.Extracting meaningful information from omics data is very important for understanding complex biological processes.The characteristic of omics data is "small N,large P",that is,large characteristic number and small sample size.Traditional feature selection methods ignore the complex interaction between each feature(biomolecule),so it is of great significance to study the interaction between features(biomolecules)from the systematic and macroscopic network level.In this dissertation,the interaction between features(biomolecules)is studied,and two methods of biomarker searching based on network are proposed.NALC(Network Analysis Method Based on LC-TSP)is proposed for biological network construction and module biomarker searching.This method studies the differences of the interaction between features on different samples,mines the information of distinguishing different samples contained in the linear combination relationship between features,and constructs the biological network.Based on the characteristics of the established biological network,a module search strategy based on key edge is proposed to determine the important sub-network in the network.The experiment on 10 public datasets shows that the classification accuracy,sensitivity and specificity of the feature subset determined by NALC method are better than those by k-TSP,LC-k-TSP,SVM-RFE and INDEED methods in most cases.A module biomarker searching method IBN(Integrated Biological Network Analysis Method)based on multi angle fusion network is proposed.In order to analyze the omics data more comprehensively and systematically and determine the key information of the problem,IBN measures the linear and non-linear relationships between features from three perspectives,establishes a fusion network based on three difference networks of LC-TSP,MIC and Spearman,and proposes a module search strategy based on the key nodes.Experiments on 10 public datasets show that the classification accuracy,sensitivity and specificity of the modules determined by IBN are better than those of the analysis methods based on a single perspective,and better than INDEED method in most cases.The two methods proposed in this dissertation are network-based module biomarker identification methods.NALC builds a biological network based on LC-TSP,and searches the important modules from the key edges;IBN integrates three biological networks to define the important modules.The experimental results on seven common datasets show that IBN method is superior to NALC method based on the linear relationship between features,but the linear combination relationship is easy to explore its biological significance.