Abstract
RNA - Protein interactions have vital roles in several cellular processes such as RNA transfer, gene regulation at the transcriptional processes and sequence encoding. RNA-binding prediction is a very important aspect of the analysis that helps in identifying the motifs that bind to DNA and for gene regulations. Predicting and recognizing the proteins that bind to RNA is a major challenging and complex process due to structural biology. Previously, several computational methods have been used and developed for predicting RNA-binding proteins (RBPs) using Support Vector Machine other than many other machine learning techniques. This paper proposes a novel computational approach for predicting RBPs using Optimum Path Forest (OPF) classifier in conjunction with the information of predicted RNA-binding residues. Moreover, the statistical information, mainly the singlet and doublet propensity, have been taken into consideration. For a given protein, its RNA-binding residues are predicted and then checked whether the protein binds to RNA or not through positive and negative samples based on the information from that prediction methodology. The results for the previous step can be classified as "Binding Protein", "Nonbinding Protein", "Binding Protein predicted as Non-Binding Protein" and "Nonbinding Protein predicted as Binding Protein", and in this case if the protein cannot be identified then the OPF classifier is used to determine the protein prediction status. The OPF classifier is used incorporated with the amino acid composition feature. The results showed that the statistical information and the binding propensity measures of the predicted RNA-binding residues especially contributed to the prediction process. In addition, the classifier has improved the overall performance of RBPs prediction process.