Research Of MicroRNA Target Predictions Based On Machine Learning Algorithm

MiRNAs(miRNAs) are a family of end endogenous single-stranded RNA around22nucleotides in length. It can lead to messenger-RNA’s to be degradation or to repress itsexpression after gene’s transcription,which will control gene expression level and playimportant roles in post-transcriptional regulatory functions. miRNAs belong to functionalRNA and act via bybridiaztion to target mRNA such as partial complementarity orcomplete complementarity. MiRNA target identification is far behind in miRNA researchin current.For further research of miRNA function, the identification of more miRNApositive targets is needed urgently. In the Bioinformatics field,the research of miRNAtarget predictions is to structure classifier model of miRNA target predictions by designingdifferent Bioinformatics algorithms.The miRNA target prediction algorithm based on rulesis confronted with drawbacks such as poor predictive accuracy and high false positiverate.MiRNA mechanism of action is not very clear by biologists.so the machine learningalgorithms attempt to intelligently judge identification rules through the statistical analysiswhich have become to a important method for the research of miRNA target predictions.Inview of the miRNA target predicitions of high dimension, nonlinear small training sampleset, The Support vector machine(SVM) algorithm is selected to sovle the problem ofmiRNA target classification and identification.And it has great significance both toimprove the performance of miRNA target predictions and identify more new miRNApositive target.Firstly,we present the SVM-MRFS algorithm,which bsed on the new SVM algorithmof v-SVMin order to build miRNA target predictions classifier model.The SVM-MRFSalgorithm give the defininition for the efficiency of features on the basis of the maximummargin classifer in the-SVMalgorithm.It sort the original feature sets in accordancewith the efficiency of features.Through the recursive training of different feature subsets,itcan search for the best feature subset and then predict the miRNA target genes.Secondly, we present the SVM-RRFD algorithm,which committed to eliminateredundant features to build miRNA target identification algorithm.The algorithm has analysis of the SVM-MRFS algorithm,which neither eliminate the redundant features nortake into account the the influence of features with smaller feature efficiency.To solve thisproblem, Based on the-SVMclassifier model the SVM-RRFD algorithm has analysis ofthe relationship between every two features in original feature set and defines the FeaturesRedundancy Standard. Through the fusion feature selection and classificationrecognition,the SVM-RRFD algorithm can find the best combinations of featuresaccording to a double standards of Features Redundancy and Features Efficiency.The bestcombinations of features can give consideration both classifier recognition performanceand generalization performance.Then,the miRNA target prediciton model with bettermiRNA target prediction performance can be built.Finally, the SVM-MRFS algorithm and SVM-RRFD algorithm are implemented inthe environment of Matlab2009a..The SVM-RRFD algorithm can obtain the best featuressubset through the iterative training.The optimal parameters combination is found by theGrid Search,and then to build the miRNA target prediction model which could better toclassify miRNA targets.