PDB Ligands: Analysing their Structure and Binding Data

The mechanism of action of the majority of therapeutic small-molecule drugs is based on formation of a non-covalent complex with a protein binding site. In spite of the availability of thousands of crystal structures of such small-molecule ligands, important aspects of the ligand binding process are still poorly understood or at least controversially discussed. These range from fundamental biophysical aspects such as ligand conformational energies to practical aspects such as the chemical identity of the ligand 3D structure deposited in, and/or perceived from, the Protein Data Bank (PDB). A similar situation exists regarding binding data, in terms of availability as well as quality of the experimental data.

The following questions need to be addressed:
- What do we have in terms of 3D structural and binding data for ligands?
- Where are these data? How accessible, how interconnected are the databases containing them?
- To what use are these data being put? What have we learned about ligand energies?
- What are the problems, and what's still missing?
- Do we need to, and can we, annotate PDB ligands with a reliability and quality score?

Ideally, we need to build a consensus on some of these questions, or at least on how to approach them in a concerted effort in order to further our understanding of protein-ligand interactions.

On 16-17 October 2008 we will hold an eCheminfo Community of Practice conference session at Bryn Mawr College, Philadelphia to address these questions related to PDB Ligands.  The session will be chaired by Marc Nicklaus (National Institutes of Health) and includes a knowledgeable panel of speakers and discussion leaders including John Westbrook (Rutgers), Howard J Feldman (CCG, Canada), Igor V. Filippov (NCI), Raul Cachau (ATP, SAIC-Frederick), Vincent T. Moy (University of Miami), Fabrice Moriaud (MEDIT, France), Paul Hawkins (OpenEye), Yulia Borodina (NCBI), Gerhard Wolber (Inte:Ligand, Austria), Marc Nicklaus (NCI), James P. Snyder (Emory), Anne Chaka (NIST), Esther Kellenberger (University of Strasbourg, France), Jim Dunbar (University of Michigan), and Janna Wehrle (NIGMS). A description of the session with presentation abstracts follows:

PDB Ligands: Analysing their Structure and Binding Data

http://echeminfo.com/COMTY_confprog08pdbligands

(Please follow continuation here to read abstracts)

Reproduction of PDB Ligand Structures by Conformational Ensembles at Different Energy Thresholds

Yulia Borodina (NCBI)

The reproduction of PDB ligand conformations by conformational ensembles obtained using low energy thresholds is compared to the reproduction by equal-sized ensembles selected randomly from the initial ensemble. The initial ensembles were created by applying a 100kcal/mol energy limit in the MMFF94s force-field, using Omega 2.2 (OpenEye Scientific Software, Inc) 3D-models and a Monte-Carlo conformer generation procedure followed by optimization of the internal coordinates while holding torsion angles fixed. The ensembles were subsequently enriched with low energy conformers by adding partly and fully optimized conformers. Optimizations and global energy minimum calculations were done using the CASE toolkit (OpenEye Scientific Software, Inc). The quality of reproduction of a PDB structure was estimated by the minimum RMSD found between any of the ensemble’s conformers and the X-ray structure (RMSD-Xray). The data set used for the study consists of ca. 2000 bioligands extracted from the PDB under the following constraints: Crystallographic resolution of 2.0Å or better, average ligand B-factor of 50Å2 or lower, and number of rotatable bonds 6 or less. RMSD-Xray averaged over compounds of equal flexibility was used as the criteria of the average reproduction quality. To validate the methodology, several test sets of ligand ensembles with conformations modeled by a force field approach within specific energy ranges were subjected to the same approach. We report on the presence or absence of significant improvement of RMSD-Xray as a function of the energy threshold in the range of 1-40 kcal/mol when compared with the same-size random ensembles, for each ligand subset of a certain flexibility range. Initial results suggest that some PDB ligand conformers have relatively high energy and cannot be reproduced by conformations created with thresholds of a few kcal/mol.

Chemically Accurate, Ultra-high Resolution X-ray Crystallography, Macromolecular Structure Analysis, and Drug Design

Raul E. Cachau, Senior Scientist, ATP/ABCC , SAIC-Frederick, Inc., National Cancer Institute at Frederick, P.O. Box B, Frederick, MD 21702, USA

Ultra-high-resolution X-ray crystallography of macromolecules (i.e. resolution better than 0.8 A˚) is a rising field that promises to provide new insight into the structure–function relationships of biomacromolecules. The number of structures solved at ultra-high resolution has considerably increased in the last 10 years due to various technical improvements, ranging from better techniques of expression and crystallization to the use of synchrotron sources for measurements of diffraction data. Below 0.8A ˚ the electron density corresponding to bonding electrons, including those of hydrogen atoms, begin to appear as deformations of the spherical density of nonhydrogen atoms. Therefore, protonation states can be experimentally determined and bond lengths and bond orders can be determined. On the other hand, ultra-high resolution structures reveal an unusually high number of deviations from standard stereochemistries, revealing the limitations of current stereochemistry dictionaries. Thus, the picture emerging from macromolecular structures at this resolution is far more complex than previously understood, requiring for its study improved tools for structure refinement, analysis and annotation. In short, macromolecular crystallography at ultra-high resolution is an unforgiving endeavor because all levels of complexity need to be handled simultaneously and in the proper manner. The lack of established protocols to do this combined analysis raises many questions concerning the results of high-resolution crystallography and to what extent the deviations from usual stereochemistries observed in high-resolution models are relevant to biological function. The implications of a positive answer to this question are far reaching as it implies our ability to unravel some of the still hidden secrets of biological function through ultra-high resolution structural analysis.

The Need for High Quality PDB Data to Drive Improvement in Ligand Binding Predictions

Anne M. Chaka, Biophysics Group, Physics Laboratory, National Institute of Standards and Technology, Gaithersburg, MD 20899, USA

Most drugs are small molecules that function by reversible non-covalent binding to proteins, modulating their activity. Once the protein target responsible for a disease has been identified, an assay is developed and experimental high-throughput screening is performed on libraries of diverse molecules resulting in the identification of a small number of active “hits”. If the structure of the protein target is known then the screening process and success rate of the screen can be dramatically enhanced through computationally “docking” potential ligands into the target’s active site and “scoring” the tightness of the binding hence utilizing a less expensive, less time consuming “virtual screen”.

Recent studies by pharmaceutical companies on the reliability of docking and scoring methodologies obtained mixed results, however (1,2). The commercial programs were in general successful in predicting the binding modes of ligands sufficiently well to aid chemical design and identify active leads in the virtual screening process. Some programs performed better than others for certain targets, but there was no clear trend to be able to predict in advance which method would be best for a given class of protein. There was, however, practically no correlation between the docking score and experimental binding affinities (IC50) (3). A NIST workshop was organized in 2006 to address how much of the discrepancy is due to computational limitations or experimental variability (4). Participants generally agreed that existing empirical scoring functions are reaching a state of diminishing returns; a deeper understanding of the physical chemistry of molecular recognition is required. More reliable estimates of binding affinities can be obtained from molecular simulation methods, but these free energy perturbation calculations are expensive with a predictive ability that is dependent upon the quality of the intermolecular potentials, or force field parameters.

Existing scoring functions for docking and molecular mechanics force fields used for dynamics simulations have been developed using an extremely limited set of available data. Functional groups that are prevalent in many drug molecules are not well represented in the data sets from which current force fields have been parameterized. This shortage of fundamental data is strongly cited as the primary reason restricting the development of accurate theory. Force field parameterization requires bond lengths and angles obtained from X-ray and microwave data, supplemented with ab initio calculations. Improvement in the description of van der Waals forces and hydrogen bonding requires contact information obtainable from high-resolution crystal structures as well as fluid properties measurements (densities, ∆Hvap).

Hence there is a critical need for a nonproprietary dataset of validated experimental binding affinities and corresponding co-crystal structures for a broad range of protein targets and their ligands (4). Such a data set would be invaluable for the testing of commercial software releases and to drive improvement in high-throughput scoring functions and force field potentials for molecular simulations and the resultant prediction of binding affinities. Current efforts and future needs will be presented.

References
1. NSF-sponsored multi-agency study “Industrial Applications of Molecular and Materials Modeling” contains detailed reports on 91 institutions, including over 75 chemical and pharmaceutical companies, plus additional data from 55 US chemical companies and 256 worldwide institutions.
2. Warren, GL, et al. “A Critical Assessment of Docking Programs and Scoring Functions”. J. Med. Chem. 2006, 49, 5912-5931.
3. Concentration of a chemical required to inhibit 50% of an enzyme’s activity.
4. NIST Workshop on “Validating Modeling and Experimental Methods to Enable Drug Discovery”, April 19-21, 2006. Organizers: AM Chaka, J. Kapur, C. Bayly, S. Brown, J. Collins, M. Gilson, M. Head, K. Merz, S. Muchmore, A. Nicholls, S. Ravichandran, A. Roitberg, F. Schwarz, S. Webb. Report available from Anne Chaka.

BindingMOAD - Analyzing High Quality PDB Entries with Known Binding Data

James B. Dunbar Jr., University of Michigan

A mining exercise was carried out to select a high quality, high precision set (HiQ) of PDB entries with curated binding data. This exercise was carried out on the BindingMOAD database and also includes data from the PDBbind database. The process of selecting this dataset will be presented as well as a preliminary analysis of some of the characteristics of this high quality set of protein/ligand complexes. The presentation begins with a detailed description of the various methods used in the HiQ set selection (i.e. diffraction-coordinate precision index) as well as notes on the issues we encountered. The analysis of the high quality set will be compared and contrasted with an earlier analysis performed on a larger set of PDB entries from BindingMOAD. The earlier set examined the differences in small molecule binding between enzymes and non-enzymes (E/NE). In addition to using the full set (in toto) comparison, we also explore the comparisons by breaking down the data into the same general subcategories of high or low affinity and enzyme or non-enzyme used in the analysis of the earlier, larger E/NE set.

Using the mmCIF Dictionary for Ligand Correction

Howard Feldman* and Paul Labute. Chemical Computing Group Inc. 1010 Sherbrooke St. W., Suite 910, Montreal, Quebec, Canada H3A2R7

The mmCIF dictionary is a curated list of all 3-letter residue codes that appear in the PDB, along with their connectivity, charges, and so on. With the recent release of the remediated PDB, all 3-letter names in all PDB files now correspond to entries in this dictionary. We have made use of the dictionary to correct ligand information in PDB files where possible, for use in Protein SILO (PSILO), a database system for managing macromolecular structures. First, a mapping between each PDB ligand and the corresponding dictionary entry is found. When all atom names match, they are used for the mapping, otherwise a graph matching algorithm is applied. Prior to PDB remediation, on a test set of 26,326 PDB files, 192,279 'ligand' residues were successfully matched to their dictionary entry (919 required graph matching), and so bond orders, charges and chirality could be copied over, while 613 could not be matched up successfully. Following remediation, on a set of 35,767 PDB files with ligands, 281,724 residues were matched to their dictionary entries, 55 required graph matching, and only 32 failed to be matched to their dictionary entry. This process assumes that the dictionary entries are correct, which is not always the case. Particularly compounds containing metal atoms or represented as multi-residue molecules are troublesome. We have made some attempts to systematically validate and correct dictionary entries but find that only very conservative changes may be made safely.

Validation using the RCSB: Good Idea or Bad Idea?

Paul C. Hawkins*, Gregory L. Warren and A. Geoffrey Skillman

Protein-ligand co-complexes from the RCSB database have been used in many studies on the quality of docking and conformer generation. However, due to the poor quality of some of the structures, many of their conclusions are invalid. This paper will discuss pitfalls associated with using structures from the RCSB for comparison or validation purposes. These pitfalls include local problems, such as poor quality fits to electron density (of ligand or protein), highly strained ligand structures and global issues such as lack of consideration of experimental error in the structural data. While nominal resolution has been frequently used for identifying good quality structures from the RCSB, much better assessments of quality can be obtained from global measures such as the diffraction-component precision index (DPI) and local measures including the real-space correlation coefficient. Consideration of these measures is mandatory when assembling a reliable set of structures for validation. Many of the problems associated with using ligand structures from the RCSB are eliminated when using small molecule crystal structures from the CSD, as there is a much greater degree of precision in these structures.

With these issues in mind, datasets for validation of conformer generation applications derived from both the RCSB and the CSD will be presented and the performance of a selection of methods on these datasets will be discussed using a number of different metrics.

PDB ligands and Drug Design: Carving out the Compound and its Cavity, and Checking out their Chemistry

Esther Kellenberger and Didier Rognan, Bioinformatics of the Drug, UMR 7175 CNRS-ULP, University of Strasbourg, F-67400 Illkirch, France

The scPDB database (1) contains the 3D structure of "druggable" protein cavities that are available in the PDB. It has been created for applications in computer-aided drug design. The screening of scPDB by serial docking of a given small molecular weight compound (or ligand) identified putative targets that were experimentally confirmed (2,3). The selectivity/promiscuity of a given ligand was investigated by structural comparison of the scPDB binding sites (4). Last, the scPDB analysis allowed us to derive empiric rules to predict cavity "hot spots", i.e. small surface patches that are likely to anchor ligands (5).

Molecular information about ligands in the PDB is necessary for both scPDB setup and usage, but depending on the purpose, different levels of accuracy are required for the ligand description.
- The binding site detection in the PDB file is based on simple ligand features (e.g. molecular type like ion, solvent, peptide..., MW, percent of buried surface...) that were deduced from the HET code and the ligand atoms characterized by their element type.
- The query of the scPDB database by ligand similarity was made possible by assigning chemically valid atom and bond types to ligands. (http://bioinfo-pharma.u-strasbg.fr/scPDB)
- For every protein/ligand complex in the scPDB, the experimental binding mode was converted into a binary fingerprint (called interaction fingerprint). The prediction of non bonded interactions can vary with the different ionization and tautomerization states of the ligand as well as with the number of polar hydrogens added to the ligand.

If the PDB remediation has greatly facilitated scPDB recent updates, ligand information is still noisy. A manual verification ensures the chemical integrity of all scPDB ligands, but the consistency of the resulting structures with published data is not systematically checked. However, we have developed strategies in drug design applications to minor errors due to mistakes in ligand typing. For example, the comparison of interaction fingerprints improves the performances of scoring in the scPDB screen by docking (6).

References
(1) Kellenberger, E et al. J. Chem. Inf. Model. 2006, 46, 717-727.
(2) Muller, P. et al. J Med Chem 2006, 49, 6768-6778.
(3) Zahler, S. et al. Chem Biol 2007, 14, 1207-1214.
(4) Schalon, C et al. Proteins: Structure, Function, and Bioinformatics 2008, 71, 1755-1778.
(5) Barillari, C.; Marcou, G.; Rognan, D. J. Chem. Inf. Model. 2008, in press.
(6) Kellenberger, E.; Foata, N.; Rognan, D. J Chem Inf Model 2008, 48, 1014-1025.

Exploration of the Chemical Diversity Generated by the Hybridisation of PDB Ligands

F. Moriaud, L. Martin, S.A. Adcock, A.M. Vorotyntsev, F. Delfaud (1) and O. Doppelt (1,2)

(1) MEDIT SA, 2, rue du belvédère, 91120, Palaiseau, France
(2) INSERM, U726, Equipe de Bioinformatique Genomique et Moleculaire (EBGM), Universite Paris 7,case 7113, 2, place Jussieu, 75251 Paris Cedex 05, France

The number and diversity of available protein/ligand complexes in the PDB are growing. Obtaining structural information on fragments complexed to a target protein is a key element and also a major limitation to the number and types of target that are amenable to fragment-based approaches. Therefore, computational methods are needed to mine efficiently all the available 3D structures of ligands complexed to proteins, both treated as a whole and as smaller fragments to increase the likelihood of fragment hopping from one target to another.

MED-SuMo[1,2] is used on a fragment database derived from the PDB protein-ligand complexes: each PDB file is converted into one or more PDB files containing the protein and a single fragment of the former ligand, with its original 3D coordinates. The fragments are stored in a MED-SuMo database as MED-Portions (500,000 entries). MED-Portions are fragments of PDB ligands annotated with dummy/bonding atoms and their 3D protein environment described by surface chemical features. A target protein surface/binding site is populated with fragments by searching with MED-SuMo for MED-Portions sharing a 3D graph of triangles of surface chemical features. Potential leads are discovered by hybridisation of MED-Portions with a de novo protocol using MED-Hybridise.

In this work, we use the MED-SuMo fragment based approach to explore the molecular diversity that can be generated by considering a diverse set of protein binding sites and by (1) populating their binding sites with fragments, (2) hybridising fragments into drug-like molecules. The protocol is able to generate valuable hybrids for a lead discovery approach. Hybrids are found to be diverse compared to the PDB: 99% of the hybrids have a scaffold which is not found in the PDB (results from a protein kinase and a GPCR application). The quality of the hybrids are assessed by computing their strained energy and the occurrence of their scaffold in Pubchem.

References
[1] Jambon M, Imberty A, Deléage G, Geourjon C “A new bioinformatic approach to detect common 3D sites in protein structures” PROTEINS: Structure, Function, and Genetics 52:137-145 (2003)
[2] Jambon M, Andrieu O, Combet C, Deléage G, Delfaud F, Geourjon C “The SuMo server : 3D search for protein functional sites” Bioinformatics Vol 21, n°20, 3929-3930 (2005)

Barry Hardy

eCheminfo Community of Practice and Research

Structural Biology eCheminfo cheminformatics chemoinformatics bioinformatics Medicinal Chemistry Computational Chemistry Virtual Screening PDB PDB Ligands Molecular Modelling Molecular Modeling pharmaceutical pharma meeting workshop training Oxford Critical Path toxicology Bursary Life Sciences Pharma Drug Discovery Research and Development Drug Development Healthcare Innovation Knowledge Management events

Intermolecular Forces of Cell Adhesion Molecules

Vincent T. Moy, Department of Physiology and Biophysics, University of Miami Miller School of Medicine, 1600 N.W. 10th Avenue, Miami, FL 33136, USA

As part of the body’s natural defense system against foreign pathogens, leukocytes enter the vascular wall and migrate to the secondary lymphoid organs or the site of infection or injury. Mediating these important immune system events are adhesion molecules including selectins and integrins that support the attachment of leukocytes to endothelial cells of the blood vessel wall. Specifically, the interaction of P-selectin and its ligand PSGL-1 facilitates the initial attachment and subsequent rolling of the 1 with theirb4a2 and bLaleukocytes, while the interactions of the integrins ligands ICAM-1 and VCAM-1, respectively, promote the arrest and firm adhesion of leukocytes. We have carried out single molecule force measurements to characterize the intrinsic dynamic properties of these adhesion complexes to better understand their role in cell adhesion. Our atomic force microscopy measurements revealed that the unbinding of the selectin and integrin complexes involves overcoming at least two activation barriers. The inner barrier of these complexes, which grants these complexes its tensile strength at high forces, is mediated by a divalent metal ion. The affinity state of the adhesion complexes are determined by the height of the outer activation energy barrier, which also determines the dissociation kinetics of the complex in the low force regime. Moreover, our measurements revealed that integrin activation stemmed from the elevation of the outer energy barrier.

High Quantum-chemical Ligand Energies - True Binding Effects or Crystallographic Artifacts?

Marc Nicklaus, National Institutes Health

We analyzed conformational changes of ligands binding to proteins in an early paper in the field (Nicklaus et al., Bioorg. Med. Chem. 3, 411-428, 1995). Both geometric and energetic aspects were studied. Surprisingly high conformational energies relative to the vacuum global energy minimum were found for some ligands. The question of such high ligand energies is still discussed controversially to this day. The previous study was necessarily limited by methodology due mainly to availability of computing power, and by number of structures due to historically much smaller database sizes. To improve on this analysis in all aspects possible, and to attempt to provide a more definitive answer to the possibility of high ligand conformational energies, a new effort is being undertaken to analyze ligand energies. To this purpose, a high-quality subset of all ligand occurrences (individual 3D coordinate sets) in the Protein Data Bank (PDB) has been compiled by filtering the entire PDB ligand database by crystallographic resolution, average ligand B-factor, true ligand nature and other criteria. Structures from this subset are then submitted to high-level quantum chemical calculations using the program Gaussian 03 to obtain energies in step-wise optimization of first bond lengths, then bond angles, and finally dihedrals. These energies are compared to the energy of global energy minimum structures obtained in molecular mechanics force field approaches, fully re-optimized at the quantum chemical level. Typical levels of theory used are Density Functional Theory (DFT) computations using B3LYP/6-31G(d) for optimizations, B3LYP/6-311++G(3df,2p) for single-point calculations. An attempt is made to discuss the obtained results as to the possibility of distinguishing whether high-energy conformations of ligands found in the PDB are true characteristics of protein-ligand complexes or artifacts introduced in one of the many steps from growing the crystal to deposition of the coordinates in the PDB.

What Energy Price Does a Drug Pay To Bind To a Protein Target?

James P. Snyder*, Ana Alcaraz, Seth Childers, Yong Jiang, Scott Johnson, Andy Prussia, Pahk Thepchatri, Suwipa Saen-oon, Jennifer Sorrells

If a molecule in equilibrium with one or more conformers in solution is characterized by a relative DG > 3 kcal/mol, its population is < 99.4% at 298 K. Chemists use this rule-of-thumb to rationalize yields and relative rates of reactions, while medicinal chemists apply it to the ligand-protein binding process. If the conformational strain of a drug or ligand conformation is much higher than 3 kcal/mol, it is believed by many that such a ligand has a low probability of binding to its macromolecular target. Over the past 10 years, several computational studies have attempted to verify the 3-kcal rule. With conflicting interpretations placing the accessible energy window somewhere between 3 and 40 kcal/mol, agreement has yet to be reached. The most comprehensive and recent study on 150 proteins complexed with drug-like ligands by Perola and Charifson (J. Med. Chem. 2004, 47, 2499-2510) suggests that global strain energies of 10 kcal/mol are common (i.e. at least 10% of ligands), while 25 kcal/mol ligand strain energy can be tolerated within protein-ligand complexes. In the present study, we examine this concept by evaluating structures and energies of both bound and “free” ligands; namely X-ray structures for the former and conformationally generated global minima for the latter. By combining molecular mechanics calculations, the fits of small molecules to X-ray crystallographic densities and NMR analysis of the conformations of ligands in solution, we conclude that it is likely that drug conformational strain energy rarely exceeds 3-5 kcal/mol in the protein binding event.

Are we there yet? Expanding the Target-Ligand Structure & Affinity Dataset

Janna Wehrle, National Institute of General Medical Sciences

Computational drug docking and screening has had notable successes identifying promising lead compounds, but it still suffers from high failure rates and inconsistent predictive power. A series of meetings among industrial and academic computational chemists and federal staff in 2005 and 2006 identified the small number of matched sets of high resolution structural and affinity data for protein-drug ligand complexes as a major roadblock. Because the nucleus of the needed dataset appears to exist, albeit in a fragmented and incomplete state, in industrial and academic laboratories across the country, it seemed that a substantial increase in high quality benchmarking data should be possible fairly quickly. As a result the National Institute of General Medical Sciences requested (http://grants.nih.gov/grants/guide/rfa-files/RFA-GM-08-008.html) and received proposals for a Drug Docking and Screening Data Resource. Its goal is to facilitate algorithm development and benchmarking for in silico drug docking and screening tools.

The Data Resource will collect under-utilized or incomplete data sets from industrial and academic groups, refine and curate data as needed, complete datasets with new crystal structures and biochemical experiments to optimize their utility, and make all results readily and publicly available. The focus will not be on massive data collection, but rather on careful design of datasets to optimize our understanding of molecular interaction potentials, leading to more reliable docking predictions. The Data Resource will be administered as a cooperative agreement, with active participation of NIH staff, especially in issues such as intellectual property, prompt release of data, and software and other resource sharing. Data housed in or developed by the Resource will be freely and conveniently available to all users and well-integrated with existing data sources, such as the Protein Data Bank.

The success of the Data Resource will depend upon the active involvement and goodwill of the computational drug development communities in academia, industry, and government. Active community consultation will be key. For example, the Data Resource could facilitate community benchmarking exercises by scheduled release of selected data sets. Review of applications will be complete and an award is expected by September 30, 2008.

Small Molecule Data in the Protein Data Bank Archive

John Westbrook (1), Kim Henrick (2), and Dimitris Dimitropoulos (2)

(1) RCSB Protein Data Bank, Rutgers, The State University of New Jersey, Piscataway, NJ, USA
(2) PDBe, EMBL - European Bioinformatics Institute, Hinxton, Cambridge, UK

The Protein Data Bank archive contains over 8350 unique chemical components. These chemical components are described in a data dictionary maintained by the Worldwide Protein Data Bank (wwPDB; wwpdb.org). In 2007, the wwPDB Chemical Component Dictionary was extended to include IUPAC atom nomenclature for standard amino acids and nucleotides, stereochemical assignments, aromatic bond assignments, experimental model and computed ideal coordinates, systematic names and chemical descriptors. Some improved web tools which take advantage of the more detailed chemical descriptions and assist users in finding small molecules in the PDB will be described.

The Application of QM/MM Refinement in Protein-Ligand Complexes for Structure-Based Drug Design

Lance Westerhoff, QuantumBio

Determining the structure of a small molecule (drug candidate or lead compound) to a biological receptor (protein implicated in disease) is a necessary step in this methodology. The dominant experimental approach used to achieve this goal is X-ray crystallography, while nuclear magnetic resonance (NMR) plays a lesser role. X-ray techniques provide astounding insights into the structure of protein-ligand complexes, but can be hampered by the resolution to which a crystal diffracts and the refinement process can be hampered by the lack of good potentials for novel small molecule compounds. An approach to improve structure quality in protein-ligand crystallography is presented by introducing the hybrid quantum mechanics and molecular mechanics (QM/MM) methods.

The QM/MM refinement leads to improvement of the local geometry with more accurate potential function for the ligand of interest taking into account interactions between protein and ligands. Benzamidine or benzamidine derivative is a good mimic for the guanidinium moiety of the arginine, which is a potential nonpeptide anatagonist of the receptors. The geometries of benzamidine show different preferences in free and bound states in terms of a single internal rotational angle about the bond connecting amidine and phenyl moieties. The force parameters of benzamidine are not accurate in the traditional refinement. The environment of the active site can alter torsional profiles dramatically. QM/MM refinement could provide preferred conformations for benzamidine with the best explanations of the density, while making reasonable contacts with different receptors. As the methodology develops and is further refined, the tool-box of structure based drug design will gain an important new method which will enable drug development for targets inaccessible to today’s mainstream drug discovery paradigm.

Structure-focused Pharmacophores for Drug Discovery from Protein-bound Ligands in the PDB

Gerhard Wolber (1,2) and Thomas Seidel (1)
(1) Inte:Ligand GmbH, Mariahilferstr. 74B/11, 1070 Vienna, Austria
(2) University of Innsbruck, Institute of Pharmacy, Center for Molecular Biosciences, Innrain 52, 6020 Innsbruck

Chemical-feature based pharmacophore models have been established as state-of-the-art techniques for virtual screening [1]. While feature-based pharmacophore recognition from a set of bio-active ligands is implemented in a number of programs [2], the recognition of 3-D pharmacophores from macromolecular complex structures with bound ligands is hardly used and has first been implemented as a fully automated procedure in LigandScout [3].

For deriving protein-ligand interaction models, the correct perception of the ligand is crucial since chemical interaction patterns are converted to 3D pharmacophore queries that represent an abstraction of the binding mode of a specific ligand. We present algorithms that automatically perceive the bound ligand structure allowing for recognition of problems in the 3D structure using a coordinate template approach for the recognition of the hybridization states and a reverse force-field interpretation for the assignment of bond orders. From interactions of the interpreted ligands with relevant surrounding amino acids, pharmacophore models reflecting functional interactions like H-bonds, ionic transfer interactions or lipophilic contacts are created and projected into 3D space. The problem of tautomers for hydrogen bonding formation and problematic assignment of double bonds will be discussed in terms of their relevance to chemical interaction patterns with the protein. Finally, geometric fitting of the automatically generated pharmacophore to the bound ligands is compared with docking methods in terms of conformational recognition, flexibility and eligibility for virtual screening.

References
[1] H. Kubinyi. In Search for New Leads, EFMC - Yearbook 2003, 14-28.
[2] G. Wolber, T. Seidel, F. Bendix, and T. Langer. Molecule-Pharmacophore superpositioning and pattern matching in computational drug design. Drug Discov Today (1-2): 23-29 (2008).
[3] G. Wolber, T. Langer. LigandScout: 3-D Pharmacophores derived from protein-bound ligands and their use as virtual screening filters J. Chem. Inf. Model., 45, 160-169 (2005).

Barry Hardy

eCheminfo Community of Practice

Structural Biology eCheminfo cheminformatics chemoinformatics bioinformatics Medicinal Chemistry Computational Chemistry Virtual Screening PDB PDB Ligands Molecular Modelling Molecular Modeling pharmaceutical pharma meeting workshop training Critical Path toxicology Bursary Life Sciences Pharma Drug Discovery Research and Development Drug Development Healthcare Innovation Knowledge Management events

Comments

[exposed skin product]

Infantile acne is a term used for the acne that affects toddlers between the age of three and 12 months. On rare occasions infantile acne may persist in toddlers until the age of three. This type of acne is slightly more severe than baby acne that affects toddlers immediately after birth. Infantile acne usually results from hormonal changes during the birth of the toddler. Toddlers affected with infantile acne generally have yellow papules on their face. They may also have blackheads whiteheads and pustules.