Predicting Protein - RNA Binding Sites Using Statistical Characters

Protein and RNA interactions play essential roles in a number of biological regulatory mechanisms. Effectively identity the binding interfaces can help understand the interaction. In this paper, we took statistical information into account, mainly the singlet propensity and doublet propensity, and added the two propensities with the sequence information, using machine learning method to predict the interfaces. Results showed that adding statistical characters can improve the prediction precision, especially the doublet propensity. Besides, we constructed three more data sets based on the protein-RNA complex function, and found out for the first time that different complex function data sets show significant differences in prediction precision.