Discovering Protein-Protein Interaction Sites from Sequence and Structure

Introduction

In the recent years, the biology community has been fascinated by snapshots of complex inner workings of cellular activities from various different angles enabled by the advancement of high throughput experiments. The construction of proteome-wide protein-protein interaction maps provided us for the first time a vivid visualization of networks of physical protein binding which maintain cellular events involving cell signaling, gene regulation, and metabolism (Giot et al. 2003;Uetz et al. 2000;Ito et al. 2001). On the other hand, genome sequencing projects and structural genomics projects continue to accumulate sequence and structure information of individual proteins. Therefore, the urgent need in the post-genomic era is to capitalize these two types of enormous amount of data to provide useful and practical information to the biology community. Effectively, protein-protein interaction sites plays a crucial role in proteomics; a greater understanding of protein-protein interaction sites ultimately bridges the gap between abstract information of protein-protein interaction (PPI) network and physical entity of protein structures, substantiating protein-protein interaction data as a critical platform for advanced molecular recognition research and experimental design. For example, knowing (or predicting) binding residues in a PPI site enables designing point mutation experiments to verify the role of the residues in the PPI network. In this chapter, we review current computational methods for predicting PPI sites from sequence and structure and extend discussion to the future direction of this field.

Physical aspects of protein-protein interactions are well studied in the context of protein-protein docking prediction. In the protein-protein docking area, at a molecular basis, these physical protein-protein binding events are commonly classified into three major groups in the context of their function: (1) enzyme-inhibitor complexes, (2) antibody-antigen interactions, and (3) other types of interactions (Hwang et al. 2008;Mintseris et al. 2005;Chen et al. 2003). This classification has been proved quite useful to atomic-level analysis employed for protein-protein binding site prediction and protein-protein docking. A genome-scale PPI network would include interactions of proteins of much broader range of functions, but here we mention these three classes to illustrate variation of the nature of protein-protein interaction.

An example structural complex for each interaction class is shown in Figure 1. The class of enzyme-inhibitor interactions involves the inhibition or regulation of an enzyme function by physical binding. It has been shown that there is a general tendency for enzyme-inhibitor interfaces to be more evolutionary conserved than other surface residues (de Vries et al. 2006;Nooren, Thornton 2003a;Nooren, Thornton 2003b). Further, antibody-antigen interactions are generally established through hydrophobic contacts, where regions of the hyper-variable loops are generated through rapid mutation resulting in the effective recognition of complementary antigenic proteins. Interactions classified as other, include obligate homo- or hetero-oligomershetero-oligomers (permanently bound) and other transient (temporarily bound) complexes involved in cell signaling, e.g. protein kinase interactions and G-coupled receptors. Classification of permanent and transient interaction is also important as the binding sites of the two classes exhibit notable differences.

Figure 1.

Examples of the different groups of protein-protein interactions: (A) PDBID: 1BRS, a Barnase-barstar complex (B) PDBID: 1KXQ, an alpha-amylase to antibody interaction, and (C) PDBID: 1B6C, an example of the other types of interactions such as those transient that involve kinase to isomerase binding