Handling Protein Flexibility in Docking and High-Throughput Docking: From Algorithms to Applications

CAVASOTTO, CLAUDIO N.

Virtual Screening: Principles, Challenges, and Practical Guidelines

WILEY-V C H VERLAG GMBH

Año: 2011; p. 245 - 262

For the last two decades, experimental high-throughput screening (HTS) of in-house or commercially available compound collections was the method of choice of the pharmaceutical industry for hit/lead identification.  However, its failure to deliver according to the expectations, coupled with the cost of HTS experiments, led to the development and use of computational methods in this task.  Thus, a variety of in silico methods rooted in physical principles and/or experimental knowledge became available for different purposes, including but not limited to the prioritization of sub-libraries of compounds for experimental evaluation.  The use of a three-dimensional (3D) structural representation of the target paved the way for structure-based approaches, such as docking-based virtual screening, structure-based lead optimization, design of target-specific virtual chemical libraries, characterization of ligand:receptor interactions at an atomic level and probing structure-function relationships among others.  It has been reported that a large amount of money could be saved in drug discovery projects if target structural information is used in the early stages of hit/lead development.  This, coupled with the growth of experimentally determined structures and the advances in homology modeling  has triggered the wide use of structure-based methods in drug discovery. If the experimental structure of a ligand:receptor complex is not available, this may be in silico modeled through ligand docking.  The sequential process of docking a virtual chemical library and assigning a ?score? to each compound -which measures the likehood of binding- is usually referred to as high-throughput docking (HTD) or structure-based virtual screening (SBVS).  The main goal of HTD is to prioritize the compound library such that only those compounds with a high chance of binding to the protein will proceed for experimental testing.  HTD also provides a 3D representation of the candidates complexed with the target. The first attempt to predict non-native ligand:receptor complexes goes back to 1976, when Beddell et al developed handmade representations of the protein structure in an attempt to design compounds that fit into hemoglobin.  Since the early 80s, when the first computer docking methods were developed , there has been an active development and improvement of docking and HTD methods.  As a necessary condition, a HTD program should incorporate a good conformational sampling of the ligand in the field of the receptor, a reliable scoring function (able to discriminate potential binders from non-binders), and should be computationally affordable.  However, an inadequate representation of the system may certainly limit the performance of docking and HTD methods.   The most common problems are uncertainty in the protonation states of ligand and residues due to induced pKa changes, inclusion/exclusion of ions and water molecules in the binding site, flipping of asparagines and glutamines, errors in assignment of histidine tautomer states, and not accounting for protein flexibility, since most docking methods were developed based on Fischer?s theory of lock-and-key, i.e., using a rigid representation of the receptor.  In this chapter we will present case studies where docking and HTD methods accounting for protein flexibility were used in the context of drug discovery/design or lead optimization projects.  This will be preceeded by a theoretical analysis of the problem, followed by concrete examples of docking failures stemming from neglecting protein flexibility.