Research On Microrna Recognition Based On Support Vector Machine

The discovery of microRNA (miRNA) is the precondition of research of miRNAregulation mechanism and development of genetic diseases related drugs. The biggestdisadvantages of experimental approaches are long cycle and high cost. The machinelearning approaches is complement to experimental approaches. Furthermore, themachine learning approaches can capture miRNAs that are expressed in specific timeor specific tissue, and miRNAs with low-expression levels. Constructing the classifierneed the features of samples as input. Little kind and number of features whichdescribe the characteristics of samples will lead to weak generalization ability of theclassifier. Large number of features will increase the complexity of model, and bringnegative impact on classification reducing the classification accuracy. Selectinginformative feature subset and proper machine learning approach can effectively solvethis problems, research of miRNA recognition based on support vector machine (SVM)is conducted in this paper.Firstly, according to the studies and reports about the sequence conservation,structure conservation and statistical regularities of miRNA precursors, we extractsequence features, structure features and structure-sequence features of miRNAprecursors as the initial feature set, then quantify the features.Secondly, for the traditional SVM to classify miRNAs treat the features equally,negative impact brought by trivial features or irrelevant features will lead to lowclassification accuracy and weak generalization ability of the model. A method ofmiRNAs recognition approach based on information gain feature selection algorithm isproposed in this paper. This algorithm keeps the features which contribute to theclassification task more as a new feature subset, and a classifier based on SVM withthis feature set is constructed.Thirdly, for the negative impact brought by irrelevant features or redundantfeatures will lead to high complexity of model and weak generalization ability of themodel, a method of miRNA recognition approach based on wrapper feature selection algorithm of genetic algorithm and SVM is proposed. This algorithm search theoptimal feature subset by the interactive way between genetic and support vectormachine, and a classifier based on SVM with this feature set is constructed.Finally, the data sets of human miRNA precursors are used to validate theproposed miRNA recognition approaches. Then the comparative analysis of existingrecognition approaches and approaches proposed in this paper is conducted. Also, weforecast next step of work.