Specific binding interactions are central to many biological processes and pathways. Similarly, most drugs act by binding specifically to a site on a target protein, thereby modulating protein activity. The quest for new drugs relies on many approaches, including computer-based docking. This computational approach has the potential to have a profound positive impact on drug discovery, justifying the keen interest in this field by pharmaceutical researchers. Over the past fifteen years, subsequent to the initial description of the DOCK program and in parallel with the exponential increase in the number of available high-resolution protein structures, many docking programs of use as virtual screening tools in the drug discovery process have emerged. Understanding the similarities and differences of different methods as well as their capabilities and limitations is both important and increasingly challenging.
The main objective of our "Virtual Screening, Docking and Scoring" program is to foster discussion amongst researchers working on both development of screening and docking methods and the application of such methods to drug discovery. This interaction will lead to a better understanding of the current state-of-the-art, improved screening and docking tools in the future, and enhanced awareness of how to apply the current set of tools.
We are holding InterAction Meeting sessions at Bryn Mawr College, Philadelphia on 11 October ’05 and in Basel, Switzerland on 9 November ’05 on Virtual Screening, Docking and Scoring. The purpose of the meetings is to bring together world-class researchers working in the computational screening and docking area, and have them present and discuss recent and new results in cutting-edge research areas relevant to drug discovery. The emphasis of the meeting activity is on communication and open discussion, which we hope will lead to new ideas and collaborations towards progress and innovation in the disease areas discussed. A significant amount of time at the meeting will be devoted to question time and panel and group discussions towards this goal.
A description of the session and invited speaker presentations with abstracts follows:
eCheminfo InterAction Meeting Session, Philadelphia, 11 October 2005
Virtual Screening, Docking & Scoring
chaired by Max Cummings (Johnson & Johnson)
eCheminfo 2005 US InterAction Meeting, 11-12 October 2005, Philadelphia, USA
Program Updates loaded in the Program Areas on http://echeminfo.com/
Invited Speakers & Discussion Leaders:
Reducing false positives in virtual screens on kinase targets, Emanuele Perola (Vertex)
New Docking Methods for Pose Prediction and Enrichment, Mark McGann (Openeye)
Molecular Docking as a Virtual Screening Tool, Renee DesJarlais (Johnson & Johnson)
Exploiting protein-specific information in docking, Willem Nissink (Cambridge Crystallographic Data Centre)
You can't find what's not there: the importance and the pitfalls of multiple representations of molecules in dockable databases, John Irwin (UCSF)
eCheminfo InterAction Meeting Session, Basel, 9 November 2005
Virtual Screening, Docking & Scoring
chaired by Miklos Vargyas (ChemAxon)
eCheminfo 2005 European InterAction Meeting, 9-10 November 2005, Basel, Switzerland
Program Updates loaded in the Program Areas on http://echeminfo.com/
Invited Speakers & Discussion Leaders:
Pattern Recognition and Grid Computing in Drug Discovery, Graham Richards (University of Oxford)
The Issue of Protein Flexibility in Docking-Based Virtual Screening, Xavier Barril (Vernalis)
Discovery of cell-permeable nonpeptide inhibitors of beta-secretase by high-throughput docking and continuum electrostatic calculations, Amedeo Caflisch (University of Zurich)
Beyond ligand flexibility: improvement and validation of rDock for Structure-Based Drug Design, David Morley (Enspiral Discovery Ltd)
Structure-based identification of GPCR ligands by high-throughput docking, Didier Rognan (University of Strasbourg)
[COMPLETE ABSTRACTS FOLLOW IN REMAINDER OF POST]
ABSTRACTS
PHILADELPHIA
Molecular Docking as a Virtual Screening Tool
Renee L. DesJarlais, Alan C. Gibbs, Venkatraman Mohan, Edward P. Jaeger, Maxwell D. Cummings, Johnson & Johnson Pharmaceutical Research & Development, L.L.C.
665 Stockton Drive, Exton PA 19341 and 8 Clarke Drive, Cranbury NJ 08512
When embarking on a screening campaign, it is often necessary to limit the number of compounds to be tested. Occasionally resources will dictate testing only a relatively small subset of available molecules. In such cases, it is important to select compounds for screening with the highest likelihood of success. Where a structure of the molecular target is available, a variety of docking tools are available for virtual screening of a compound collection. It is difficult to find unbiased assessments and comparisons of these docking methods as virtual screening tools. We sought to do our own tests with a variety of docking methods on a variety of targets. We seeded a 1000 molecule database of drug-like molecules from the MDDR with known actives for 5 different targets. The performance of six different programs was evaluated based on the ability of each program to rank actives well, the correlation of scores with binding affinity and the ability of each program to find modes consistent with crystallographically determined structures. Detailed analysis of the results will be presented, and some specific challenges of this type of study will be discussed.
New Docking Methods for Pose Prediction and Enrichment
Mark McGann, PhD, OpenEye Scientific Software, 222rd Street Suite 3211, Cambridge, MA 02142, USA
The two basic jobs of a docking program are determining the correct position of ligands within the active site and correctly ranking the relative activities of those ligands. In terms of the former, many docking programs can reliably produce a set of potential poses in which at least one pose can be considered correctly docked. Reliably picking this single correct pose out of the entire set with a scoring function is often more problematic. This presentation describes a new method of picking out the correctly docked pose using multiple scoring functions. This "consensus structure" method is distinct from what is generally called "consensus scoring" because the former is used to determine the docked structure while the latter is used for enrichment. While the "consensus structure" method is a structure determination method, it also indirectly improves enrichment by virtue of its improved pose prediction. This presentation will also detail the use of MASC to more directly improve enrichments. The MASC method attempts to correct for systematic errors in any scoring function by comparing a ligand's score to that ligand's score in a set of reference protein targets. A corrected score is then assigned based upon how much better or worse the ligand scores in the current target relative to the reference targets. Conceptually this is equivalent to calculating the average delta G of moving a ligand from the reference targets into the current target, rather than calculating the delta G of moving the ligand from solvent into the current target.
Exploiting protein-specific information in docking
Willem Nissink, Cambridge Crystallographic Data Centre, 12 Union Road Cambridge, Cambs, CB2 1EZ , United Kingdom
State-of-the-art software docking tools nowadays allow the user to dock ligands with reasonable confidence. Still, it is clear that our scoring functions are not perfect, and often overlook weak interactions like CH...O hydrogen bonds completely. Neither is it straightforward to take into account binding site information, although such knowledge can be quite obvious for individual targets. Examples of the latter are e.g. knowledge about displaceable mediating waters; placement of a common ligand scaffold; metal coordination geometries, or presence of flexible side-chains.
In this talk we will discuss developments that enable easy use of additional knowledge, either during docking, or, a posteriori, through post-processing tools. We will show the approach used for inclusion of water information in GOLD, and present results for the automatic assessment of water-mediated binding modes during docking. We will show improvements to the handling of metals, and docking restraints. Further we will introduce our newly developed post-processing tool, SILVER, that allows the user to set up target-specific filters using a variety of contact- and hydrogen-bond descriptors.
Reducing false positives in virtual screens on kinase targets
Emanuele Perola, Vertex Pharmaceuticals, 130 Waverly Street, Cambridge, MA 02139, USA
Virtual screening is an increasingly popular approach for the identification of new chemical classes at the onset of drug discovery programs. Powerful hardware and improved software, combined with a better understanding of protein/small molecule binding, have made this technology a viable complement and in some cases a reasonable alternative to high-throughput screening. A number of reports describing successful examples of virtual screens have been published in recent years. However, while the identification of a few compound classes displaying the desired activity is generally accepted as a criterion for success, the reported hit rates usually range from 3 to 10%. Even in the best-case scenarios the majority of the high-ranking compounds turn out to be inactive. It is clear from this perspective that false positives remain the biggest challenge virtual screening practitioners currently face. In this talk a detailed study addressing the causes behind the generation of false positives in structure-based virtual screening will be described.
Structure-based virtual screens were performed on a number of kinase targets using libraries with associated high-throughput screening data. The models generated for the top-ranking compounds were analyzed and key features differentiating false positives from true binders were identified. Implementation of the acquired knowledge in the virtual screening procedure produced significant enhancements of the enrichment factors obtained on each target. On this basis a general protocol for virtual screening on kinase targets was defined, and useful insights extendible to other target classes were derived.
You can't find what's not there: the importance and the pitfalls of multiple representations of molecules in dockable databases
John Irwin, Pharmaceutical Chemistry, UCSF, 1700 4th St, Suite 501D, San Francisco CA 94143-2550 USA
The ZINC database of purchasable compounds for virtual screening is built largely from supplier catalogs available in 2D SDF format. To prepare a database for docking, we generate a 3D structure, and consider protonation, charge, and tautomeric form. If there is stereo- or regio-chemical ambiguity in the source file, that must also be addressed. One approach to uncertainty is to make an educated guess at a single representation. The approach we have taken in ZINC is to prepare multiple forms of the molecule that may be relevant. The downside to such sampling is one may create new decoys - docking hit lists may be overwhelmed by high scoring but physiologically irrelevant molecular representations, presenting new challenges to scoring functions. This talk will enumerate some of the choices to be made, describe our own strategy in some detail, and discuss examples that illustrate the advantages and disadvantages of using multiple representations of molecules in dockable databases.
BASEL
The Issue of Protein Flexibility in Docking-Based Virtual Screening
Dr. Xavier Barril, ICREA and Departament de Fisicoquimica, Facultat de Farmacia, Universitat de Barcelona, Av. Diagonal 643, 08028 Barcelona, Spain
The binding sites of pharmacological targets tend to present a high degree of plasticity, which can range from relatively small ligand-induced fit effects to major changes in the shape and molecular recognition properties of a cavity. In pharmaceutical research, this can be exploited to identify classes of inhibitors with different interaction patterns and, potentially, very different overall properties. Nevertheless, for practical, computational and theoretical reasons, docking-based virtual screening (VS) generally uses only one conformation of the receptor. Although this should be a major limitation of the method, we have recently shown that using multiple X-ray structures in VS applications is likely to lead to worse results than standard rigid-receptor docking (J. Med. Chem., 48:4432, 2005). Here we will use the Hsp90 case to discuss which and how many cavities should be selected to attain optimal VS results, both in terms of enrichment factors and chemical diversity. The use of the consensus cavity scoring method will also be discussed.
Discovery of cell-permeable nonpeptide inhibitors of beta-secretase by high-throughput docking and continuum electrostatic calculations
Danzhi Huang (1), Urs Luethi (2), Peter Kolb (1), Karin Edler (1,2), Marco Cecchini (1),
Stephan Audetat (1,2), Alcide Barberis (2), and Amedeo Caflisch (1)
Affiliation:
1Department of Biochemistry, University of Zurich, Winterthurerstrasse 190
CH-8057 Zurich, Switzerland
2ESBATech AG, Wagistrasse 21, CH-8952 Zurich-Schlieren, Switzerland
A fragment-based docking procedure followed by substructure search was used to identify active-site beta-secretase inhibitors from a composite set of about 300,000 and a library of nearly 180,000 small molecules, respectively. After docking the compounds were ranked according to the binding affinity evaluated with a linear interaction energy approach with continuum electrostatics (LIECE). Values of EC_50 lower than 10 micromolar were measured in at least one of two different mammalian cell-based assays for 12 of the 72 purchased compounds. In particular, a phenylureathiadiazole derivative (MW=322 g/mol) and a diphenylurea derivative (MW=419 g/mol) are very promising lead compounds for beta-secretase inhibition to treat Alzheimer's disease. It is important to note that for almost all of the 12 compounds, for which an EC_50 value could be measured, the discrepancies between the LIECE predicted affinity and the experimental value is within the accuracy of the LIECE approach of about 1 kcal/mol.
Beyond ligand flexibility: improvement and validation of rDock for Structure-Based Drug Design
David Morley, Enspiral Discovery, The Old Police Station, Priory Lane, Royston, Herts. SG8 9DU, UK
The use of molecular docking for virtual screening of large libraries has gained rapid acceptance in recent years, with a plethora of tools available based on a variety of scoring functions and search algorithms. However, the original paradigm on which most of the high-throughput methods are based, that of docking a semi-flexible ligand into a near-rigid receptor in the absence of explicit bound waters, has severe limitations in many cases. More recent challenges are to deal with protein and water flexibility in a meaningful way, whilst maintaining throughput, to achieve greater accuracy in binding mode prediction and ranking of ligand binding affinities.
This presentation will describe recent improvements to the rDock virtual screening platform (Vernalis, UK) to improve the ability of detailed docking studies to inform an experimentally-driven Structure-Based Drug Design (SBDD) process. In particular,
i.the stochastic search engine employed by rDock has been generalised to support protein, ligand and explicit solvent degrees of freedom.
ii. External restraints derived from experimental methods can be applied to bias the search in a variety of ways.
Validation metrics will be presented where appropriate to highlight the new features.
Pattern Recognition and Grid Computing in Drug Discovery
Prof. W. Graham Richards, Chairman, Department of Chemistry, University of Oxford, Oxford, UK
Discovering small molecules that interact with protein targets identified by structural genomics, proteomics and bioinformatics will be a key part of future drug discovery efforts. Computational screening of drug-like molecules is likely to be valuable in this respect; however, the vast number of such molecules makes the potential size of this task enormous. Here, I describe how massively distributed computing using screensavers has allowed databases of billions of compounds to be screened against protein targets in a matter of days.
With the sequencing of the human genome essentially completed, attention has turned towards proteomics and structural genomics. These experimental studies, together with bioinformatics, are producing numerous protein structures, which are invaluable starting points for drug discovery. A major future effort will be to find small molecules that can interfere with the actions of these proteins or with interactions between them – possible drugs.
The role of computational methods in this endeavour will have several aspects. First, the appropriate binding site on the protein target must be identified. Small molecules can then be screened by predicting the binding free energy of each individual compound to the site, preferably allowing some flexibility in the shape of both binding partners. But if billions of compounds are to be screened in this way, the computational power that is required becomes a limiting feature. To address this issue, we have exploited massively distributed computing – using screensaver time on personal computers (PCs) all over the world – to screen 3.5 billion compounds against protein targets that are involved in the pathogenesis of cancer, anthrax and smallpox.
Structure-based identification of GPCR ligands by high-throughput docking
Didier Rognan, Gilles Marcou and Esther Kellenberger, CNRS UMR7081, F-67400 Illkirch, France
G-Protein Coupled Receptors (GPCRs) constitute one of the most relevant family of macromolecular targets among the human genome. Until recently, most rational approaches to design GPCR ligands have been based on the knowledge of known ligands. The high-resolution X-ray structure of a GPCR (bovine rhodopsin) has opened new opportunities for a structure-based approach. Herewith, we present some good practices in GPCR modelling and subsequent ligand docking for maximizing the chances to discover new ligands. Optimizing the in silico screening against GPCR models may indeed take place at several steps of the screening process which will be illustrated by the presentation of a few successful scenarios.
Barry Hardy
eCheminfo Community of Practice Manager
Douglas Connect, Switzerland
+41 61 851 0170 (office)
eCheminfo pharma Drug Discovery cheminformatics Molecular Simulation Virtual Screening conference meeting science chemistry events
Comments