Prediction of Conformational Epitopes by Geometrical Affinity and Chemical Propensity Approaches

A conformational epitope is composed of several discontinuous segments as antigenic determinants which are spatially close to each other in the three dimensional structure. These segments form the antigen which may bind with a specific receptor of the immune system, and play an important role in vaccine designs and immuno-biological experiments. Though there are two major types of epitopes: linear and conformational epitopes, it has been estimated that more than 90% of B-cell epitopes depend on nonsequential amino acids and are geometrically clustered due to molecular folding. Therefore, prediction of conformational epitopes rather than linear ones becomes an important and challenging task for practical applications. In this paper, a novel conformational epitope prediction system was developed based on the characteristics of surface rate analysis of side chain atoms, distribution of surface curvature attribute, and physicochemical propensity of each surface residue. It is the first conformational epitope prediction system based on the combinatorial features of curvatures and surface rates of side chain atoms. In this paper, benchmark datasets were employed to train the optimal parameter settings, and thirty extra antigen-antibody complexes from three different data resources with verified conformational epitopes were adopted to evaluate the prediction accuracy. Comparing with those well-developed tools, our proposed method outperformed the others in both aspects of accuracy and efficiency. For this testing dataset, the proposed system achieved an average sensitivity of 39.4%, an average specificity of 91.2%, and an average AUC value of 0.69.