Causal Orientation-based Gene Regulatory Network Construction Algorithm Research And Implementation

Gene regulatory network construction and regulatory relationship discovery aresignificance for research of special transcription regulatory mechanism. And they areimportant work to promote bioinformatics and system biology research. Traditionalexperiment verification is time-consuming, expensive and requires antibodies for eachtranscription factor. Using gene expression data and other bioinformatics dataconstruct gene regulatory network or inference gene relationship by machine learningand statistics method can reduce experimental scale, and guide experimentverificationGene regulatory mechanism is complex and involves interaction between variouskinds of molecule. There have many model and methods has been used for generegulatory network construction and gene regulatory relationship discovery. Thisthesis study gene regulatory network inference combine with causal-orientationalgorithms and research method. And proposes causal-orientation based generegulatory network construction method and regulatory relationship predictionmethod. Detail research content as follows:(1) Review current model and methods of gene regulatory network construction andregulatory relationship inference. Compare these model and methods and Analysisthe main problems: multi-factor regulatory and statistics significance. This thesisthink feature selection technology and supervised learning method can solve theseproblems effectively.(2) Introduce causal-orientation model and methods. These model and methods canbe used for determine the orientation of regulatory relationship. Use the referenceof causal-orientation methodology we can develop method of gene regulatoryrelationship inference.(3) Inspired by idea that causal-orientation algorithm can orient gene regulatoryrelationship, we propose ANM (additive noise model) based gene regulatoryconstruction algorithm. Using this algorithm measure degree of causal-effectrelationship. First, we extend ANM based causal orientation algorithm to afeature selection algorithm. In the experiment of three DREAM5dataset, it hashigher prediction accuracy comparing with other similar algorithms. We designedand developed a tool based on this algorithm, this tool used for filter transcriptionfactors of target gene.(4) Supervised learning method can train prediction model using known regulatory relationship. We propose a supervised learning algorithm for this. Main workincluding feature extract and sampling based on imbalanced rate. We use excellentrandom forest algorithm for training predict model. Then make experiment andcomparing with non-supervised algorithm-CLR and supervised algorithmSIRENE, result shows supervised method has an advantage over non-supervisedmethod and supervised algorithm proposed by us superior to SIRENE.