Construction Of Gene Regulatory Networks And Research On Function Prediction

MiRNAs are endogenous RNAs which can't encode proteins.The length of miRNAs ranges from 21 to 25 nucleotides.MiRNA repress gene expression by binding to the 3'untranslated region of target genes,leading to target gene translational repression or direct cleavage to regulate the life process of organism.Understanding the intrinsic mechanisms of gene regulatory networks is important to understand the function of miRNA in many biological and metabolic processes.Therefore,predicting the relationship between miRNA and target genes for biology research is the research hotspot in recent years.In order to reach this goal,this dissertation presented a network construction method called PSO-GA-RS.The method was proposed to combine particle swarm optimization(PSO)with genetic algorithm(GA).And the objective function of the method is based on the weighted sum between dependency degree of rough set and the number of selected features.Firstly the sequence data from mi RNA and mRNA were used in this method and 48 dimensional features including position-based features,structure features and free energy features was extracted.Secondly,PSO and GA were combined considering the different advantages of themselves.The crossover and mutation operator of GA were introduced into PSO to find better candidate feature subsets.The optimal feature subsets was obtained by fitness function,which based on the weighted sum between dependency degree of rough set and the number of selected features.SMOTE method was used to solve the imbalance problem of samples in the experiment.Finally,the regulatory relationships between miRNA and target genes was predicted by SVM.Through doing the experiment on Arabidopsis thaliana and Oryza sativa dataset,results showed the method can predict the regulation between miRNA and target genes accurately.MiRNA plays a key role in the response to biotic and abiotic stresses.This dissertation proposed a method to predict the function of miRNA.The functional similarity matrix between miRNAs was calculated based on weighted protein networks and some correlation calculation algorithms of graphs.Then a simple bust robust ranking-based approach was used to construct miRNA functional similarity network.Finally a transductive multi-label classification(TRAM)algorithm is applied to predict miRNA function.The experimental results showed that the proposed method can predict the function of miRNA effectively on Arabidopsis thaliana and Oryza sativa dataset.