Virtual Screening on Homology Models

Abstract

IntroductionSetting up a structure-based virtual screen requires three basic steps. In addition to the optimization and selection of the docking and scoring protocol, it needs a preconditioned database of screening compounds and the three-dimensional structure of the target. Although the first two requirements could be fulfilled in virtually all discovery programs, the availability of an atomic resolution structure of the target would be a limiting factor. Virtual screening applications typically utilize structures obtained by experimental and theoretical methods. The present practice prefers X-ray structures to theoretical models; however, X-ray crystallography also has some limitations. Crystallization and X-ray crystallography of membrane proteins are very challenging tasks. The Protein Data Bank (PDB) consists of 551 coordinate sets for 202 unique membrane proteins including proteins of same type from different species (http://blanco.biomol.uci.edu/Membrane_Proteins_xtal.html (August 22, 2009)). Recent years have seen significant achievements in crystallization and structure determination of several pharmacologically important targets and antitargets. A couple of G-protein-coupled receptor (GPCR) structures have been published [1], the structure of amino acid transporters (e.g., [2]) and ion channels (e.g., [3]) have been revealed, mammalian cytochrome P450 structures have been investigated experimentally [4], and, more recently, the atomic resolution structure of the Pgp transporter became available [5]. Despite the exponential increase in the number of unique structures during the past 10 years, membrane protein structures represent only about 1% of the PDB. The overall quality of X-ray structures should also limit their applications in virtual screening. Davis et al. demonstrated [6] that high-resolution structures (<1.5 A ) are 95% based on observed data; however, structures with resolution lower than 2.5 A are rather subjective. Very recently, Nielsen and coworkers also concluded that structural artifacts such as crystal contacts and water-mediated contacts in protein-ligand complexes influence the performance of scoring functions and consequently the outcome of virtual screening [7].In the lack of X-ray structures, homology models are typically used for virtual screening. Homology models are obtained by comparative modeling using structural Virtual Screening. Edited by C. Sotriffer