Local Protein Surface Patch Method for Protein-Ligand Binding Prediction

Abstract

Characterizing the protein structures of an unknown function is an important task in bioinformatics. Function of a protein, specifically, the type of ligand that bind to a protein, can be predicted by finding similar local surface regions of known proteins. Unlike existing methods that compare the global characteristics of a protein fold, we propose a local surface-based approach which can find functional similarity between non-homologous proteins. In the proposed pocket comparison method, the pockets are segmented to surface patches, which are then compared using a modified weighted bipartite matching algorithm. The 3D Zernike descriptors, which have been found to be successful in representing protein global surface properties (Sael L, Li B et al. Proteins, 2008; Sael L, La D et al. Proteins, 2008) and global pockets shape comparison (Chikhi R et al. Proteins, 2010), are used to encode the geometric and physicochemical properties of the surface patches. By representing a pocket by a set of local patches, local similarity of binding pockets can be captured. This is effective when pocket shapes are slightly different due to flexibility of ligand molecules. The binding ligand prediction performance was evaluated on a data set of 100 non-homologous proteins that bind to either one of nine types of ligands. 84.0% of the binding ligands were predicted correctly within the top three scores using the shape and pocket size information, which is better than the previous method which uses the surface of whole pocket. The performance was further improved to 87.0% when surface properties, i.e. electrostatic potential and hydrophobicity, were added. Overall, we show that proposed method is powerful in predicting the type of ligand a protein binds even in the absence of homologous proteins in the database.