In this study, we designed two discriminative structure-derived features, namely residue electrostatic surface potential and triplet interface propensity, to characterize a protein residue together with other commonly used descriptors. A comprehensive analysis of the two newly designed features from different aspects demonstrated that the two features have excellent discriminative power on a large dataset and may reflect the underlying mechanisms of RNA-protein interactions. To incorporate information from neighbor residues to determine the RNA-binding properties of each target residue, the optimal patch type and patch size for different features are searched, and by using the searched optimal patch type and patch size for each used feature, a random forest classifier is developed and implemented in the web server RNAProSite. From the results of a fivefold cross validation on a training set and the prediction performance on a test set, we concluded that our method can predict RBRs with results better than or comparable to those of the existing approaches and could assist researchers in performing more targeted assays.
Copyright © 2020 Prof. HongLin Li's Group, School of Pharmacy, East China University of Science & Technology · All Right Reserved.