Prediction of Mammalian microRNA binding sites using Random Forests

In biological systems, microRNAs involve in the regulation of their target genes by degrading the targeted binding mRNAs or by repressing the corresponding protein products. MicroRNAs are shown to play important roles in numerous biological processes, such as disease formation, especially in cancer. Predicting the target binding site of microRNA can help to identify the novel miroRNA target genes. Either microRNAs or its target genes can be recognized as biomarkers for diagnosis of diseases, prediction of prognosis, or even therapy decision. In this study, first we apply support vector machines (SVMs), neural networks and decision tree-based approaches to select a set of useful features, which represent important characteristics for the determination of the interaction between microRNA and its target binding mRNA sequence. Next, these selected features are used in two classifiers, SVM and Random Forests, to perform prediction of microRNA target sites. The features that are selected by Random Forests itself exhibit the best performance for predicting the binding site of microRNA. Its prediction accuracy can reach about 75%.