Prediction of RNA-binding residues in protein sequences using support vector machines

Understanding the molecular recognition between RNA and proteins is central to elucidation of many biological processes in the cell. Although structural data are available for some protein-RNA complexes, the interaction patterns are still mostly unclear. In this study, support vector machines as well as artificial neural networks have been trained to predict RNA binding residues from five sequence-derived features, including the solvent accessible surface area, BLAST-based conservation score, hydrophobicity index, side chain pK(a) value and molecular mass of an amino acid. It is found that support vector machines outperform neural networks for prediction of RNA-binding residues. The best support vector machine achieves 70.74% of prediction strength (average of sensitivity and specificity), whereas the performance measure reaches 67.79% for the neural networks. The results suggest that RNA binding residues can be predicted directly from amino acid sequence information. Online prediction of RNA-binding residues is available at http://bioinformatics.ksu.edu/bindn/