Prediction of protein secondary structure using large margin nearest neighbor classification

Abstract

Prediction of protein secondary structure from a primary sequence plays a critical role in structural biology. In this paper, we introduce a novel method for protein secondary structure prediction by using PSSM profiles and large margin nearest neighbor classification. Although the PSSM profiles and traditional nearest neighbor (NN) method can be directly used to predict secondary structure, since the PSSM profiles are not specifically designed for protein secondary structure prediction, the NN method could not achieve satisfactory prediction accuracy. To addressing this problem, we use a large margin nearest neighbor model to learn a Mahalanobis distance metric via convex semidefinite programming for nearest neighbor classification. Then, an energy-based rule is invoked to assign secondary structure. Tests show that, compared with other NN methods, significant performance improvement has been achieved with respect to prediction accuracy by the proposed method.