Protein function annotation from sequence: prediction of residues interacting with RNA

Motivation: All eukaryotic proteomes are characterized by a significant percentage of proteins of unknown function. Computational function prediction methods are therefore essential as initial steps in the function annotation process. This article describes an annotation method (PiRaNhA) for the prediction of RNA-binding residues (RBRs) from protein sequence information. A series of sequence properties (position specific scoring matrices, interface propensities, predicted accessibility and hydrophobicity) are used to train a support vector machine. This method is then evaluated for its potential to be applied to RNA-binding function prediction at the level of the complete protein. Results: The 5-fold cross-validation of PiRaNhA on a dataset of 81 RNA-binding proteins achieves a Matthews Correlation Coefficient (MCC) of 0.50 and accuracy of 87.2%. When used to predict RBRs in 42 proteins not used in training, PiRaNhA achieves an MCC of 0.41 and accuracy of 84.5%. Decision values from the PiRaNhA predictions were used in a second SVM to make predictions of RNA-binding function at the protein level, achieving an MCC of 0.53 and accuracy of 76.1%. The PiRaNhA RBR predictions allow experimentalists to perform more targeted experiments for function annotation; and the prediction of RNA-binding function at the protein level shows promise for proteome-wide annotations.