A pinboard by
Wenhao Jin

PhD student, National University of Singapore


protein-protein interaction networks

RNA-binding proteins (RBPs) play an important role in post-transcriptional gene regulation (PTGR) that controls the maturation, transport, stability and degradation of RNAs. Discovering novel RBPs and recognizing the RNA-binding potential of known proteins will help increase our understanding of the PTGR system and each protein’s function. There have been many attempts to computationally predict binding of a protein to RNA, using intrinsic protein characteristics such as sequence and structure. However, extrinsic protein characteristics such as the identities of interacting protein neighbors, have not been examined as features useful for predicting RNA binding activity. In this study, we construct a classifier for recognizing RNA binding proteins utilizing both the extrinsic and intrinsic properties of proteins. We recently demonstrated the predictive power of protein-protein interaction (PPI) networks for predicting RNA binding proteins. Here, we extract RNA-binding features from protein sequences with convolution neural networks. Next, a group of accurate RBP classifiers (SONAR+) is obtained by integrating the information of PPI neighbors and sequence features with deep neural network and SVM models. We find SONAR+ outperforms other classifiers and we have experimentally validated several candidate RBPs which were previously uncharacterized for RNA interaction. SONAR+ accurately and efficiently expands the list of RBPs.