The Prediction Methods Of Protein-RNA Interactions

Protein-RNA interactions always played a key role in the cellular functions.Many critical cellular processes,such as cell motility,material transport,chromosome replication,transcription and translation,signal transduction and so on,are essential for protein-RNA interactions.With the increase of the experimental data of protein-RNA complexes in recent years,the prediction of protein-RNA interactions becomes an urgent and important issue at present.The prediction methods of protein-RNA interactions have been mainly divided into two categories: one is the computational means,and the other is the experimental means.There are many disadvantages in predicting the interaction by using experimental means,time consuming and laborious,so the computational means are welcome for predicting the protein-RNA interactions.For predicting the protein-RNA interactions by utilizing the computational means,the key problem lies in the construction of classifier models.The performance of classification is different for different classifiers.In order to further understand the classification performance of different classifiers,this paper first summarizes the advantages and disadvantages of classification performance by analyzing and comparing the three classifiers which are na?ve bayesian classifier,support vector machine(SVM)classifier and random forest(RF)classifier.Then,this paper identifies the protein and RNA interactions only based on sequence information which generally exists in the related databases.In this paper,the prediction model of protein-RNA interactions is constructed based on the sequence information.Utilizing the total interaction propensity of amino acid triplet and nucleotide acid computed from 3149 protein-RNA interaction pairs in PDB,we define a weighted interaction propensity measure,which is used to compute the individual interaction propensity of amino acid triplet and nucleotide acid for single protein and RNA pair.Then construct prediction model based on these individual propensity measures,the amino acid triplet component features and nucleotide component features.In order to avoid the feature redundancy,the prominent features are selected by utilizing the feature selection method.By comparision of several methods,the computational results prove the effectiveness and feasibility of predicting model(SVM model and RF model)and algorithm.Predicting the same dataset by using the same prediction methods,the effective results further prove that our individual propensity measures of amino acid triplet and nucleotide acid play an important role in the prediction of protein-RNA interactions.