Small world network strategies for studying protein structures and binding

Abstract

Small world network concepts provide many new opportunities to investigate the complex three dimensional structures of protein molecules. This mini-review explores the published literature on using small-world network approaches to study protein structure, with emphasis on the different combinations of descriptors that have been tested, on studies involving ligand binding in protein-ligand complexes, and on protein-protein complexes. The benefits and success of small world network approaches, which change the focus from specific interactions to the local environment, even to non-local phenomenon, are described. The purpose is to show the different ways that small world network concepts have been used for building new computational models for studying protein structure and function, and for extending and improving existing modelling approaches.

Keywords: Small world network; protein structure; protein-ligand binding; protein-protein binding.

Figure 1

Different representations of 3D protein structure provide different types of understanding. (a) The electron density map for ligand and surrounding binding site residues shows a good quality, rigid model. (b) Typical computational chemistry view, showing that protein-ligand binding involves many short hydrogen bonds. (c) A network view of all favourable polar interactions in the binding site region shows how the protein-ligand hydrogen bonds are parts of highly connected local environments, which also involve numerous bound water molecules. (d) Secondary structure view, suited for structural bioinformatics analysis. Protein is Neuraminidase in complex with Zanamivir. Images generated using PyMol (www.pymol.org), PDB Ids 2cml and 1nnc.

Figure 2

Simple chemical graph showing the three most widely used network descriptors for nodes, with the top five values for each. Node 28 has the highest degree, that is, the most connections (and is therefore a hub). Node 10 has the highest betweenness, which is a function of the fraction of shortest paths through a node (removing this node creates the two largest disconnected fragments). Node 1 has the highest closeness, that is, the inverse of the average of the shortest paths to all other nodes. This particular graph has a low clustering coefficient - no pair of connected nodes is connected to the same third node (no triangles). The characteristic path length is low (near 5) though this is not particularly meaningful as it is only a very small network.