Real Value Prediction of Solvent Accessibility by Using the k-Nearest Neighbor Method

Prediction of a protein residue solvent accessibility gives useful guidance for the prediction of the full three-dimensional structure and function of a protein. Although solvent accessibility is a continuous quantity, it has been predicted mostly by assigning each residue into one of two or three classes by using various threshold values. Predictions of real values of solvent accessibility have been attempted only recently, but the k-nearest neighbor method has never been used. In this work, we apply a simple implementation of the k-nearest neighbor method to the real-value prediction of solvent accessibility, using PSI-BLAST profiles as feature vectors and obtain results comparable to those obtained by using other methods.