 |
|
|
|
The importance of dynamics in understanding the selectivity of COX inhibitors
Philip W. Fowler and Professor Peter V. Coveney
Centre for Computational Science, Department of Chemistry, University College London, Christopher Ingold Laboratories, 20 Gordon Street, London WC1H 0AJ, United Kingdom
We present results from a classical molecular dynamics investigation into the the differences in dynamical behaviour between the COX-1 and COX-2 monotopic enzymes. Designing small molecules that selectively inhibit only one of these pharmaceutically important isozymes is a very difficult problem due to their highly similar (static) structures [1]. We believe that studying the differences in dynamical behaviour will yield significant insight and aid future non-steroidal anti-inflammatory drug (NSAID) development. This has gained importance recently following the withdrawal of Vioxx, one of the first 'COX-2 inhibitors', due to concerns about thrombotic side effects.
To the best of our knowledge, these are the first fully-atomistic classical molecular dynamics simulations (ca. 100,000 atoms) of solvated, membrane-associated COX enzymes. Previous studies made simplifications e.g. focussing on the foot of the enzyme or not solvating the protein [2, 3]. Membrane-associated apo COX-1 and COX-2 enzymes are evolved for ca. 10ns and we describe the range of dynamical effects observed. These range from conformational changes of the entire protein to rearrangements in the amino acids blocking the entrance to the active site. Both enzymes are shown to become well-associated with the lipid bilayer during the course of the simulations. It is hypothesised that the presence of the membrane stablises the enzyme - this is tested by the comparison of membrane-bound simulations with membrane-free simulations. The entry of the substrate, arachidonic acid, into COX-1 is briefly studied and we also outline our chosen classical molecular dynamics algorithm, NAMD2.5 [4], before discussing the computational requirements for our study. Finally, we discuss our plans for future work.
References:
1. R.Michael Garavito and Anne M. Mulichak. The structure of mammalian cyclooxygenases. Annu. Rev. Biophys. Biomol. Struct., 32:183-206, 2003.
2. Mafalda Nina, Simon Bernèche, and Benoît Roux. Anchoring of a monotopic membrane protein: the binding of prostaglandin H2 synthase-1 to the surface of a phospholipid bilayer. Eur Biophys J, 29:439-454, 2000.
3. Ferenc Molnar, Lawrence S. Norris, and Klaus Schultern. Simulated (Un)binding of Arachidonic Acid in the Cycloxygenase Site of Prostaglandin H2 Synthase-1. Progress in Reaction Kinetics and Mechanism, 25:263-298, 2000.
4. Laxmikant Kalé, Robert Skeel, Milind Bhandarkar, Robert Brunner, Attila Gursoy, Neal Krawetz, James Phillips, Aritomo Shinozaki, Krishnan Varadarajan, and Klaus Schulten. 'NAMD2: Greater scalability for parallel molecular dynamics'. Journal of Computational Physics, 151:283-312, 1999.
Less is More - Proteins in 2D
Birgit Albrecht, Guy H. Grant, W. Graham Richards
Department of Chemistry, University of Oxford, Central Chemistry Laboratory, South Parks Road, Oxford, OX1 3QH, UK
Protein similarity estimations can be achieved using reduced dimensional representations and we describe a new application for the generation of two-dimensional maps from the three-dimensional structure. The code for the dimensionality reduction is based on the concept of pseudo-random generation of two-dimensional coordinates and Monte Carlo-like acceptance criteria for the generated coordinates.
A new method for calculating protein similarity is developed by introducing a distance dependent similarity field. Similarity of two proteins is derived from similarity field indices between amino acids based on various criteria such as hydrophobicity, residue replacement factors and conformational similarity, each showing a one factor Gaussian dependence.
Results on comparisons of misfolded protein models with datasets of correctly folded structures show that discrimination between correctly folded and misfolded structures is possible. Tests were carried out on five different proteins, comparing a misfolded protein structure with members of the same topology, architecture, family and domain according to the CATH classification.
References:
Birgit Albrecht, Guy H. Grant, and W.Graham Richards, Protein Eng. Des. Sel. 2004; 17(5): p. 425-432
MACiE - The Classification and Computer Representation of Enzyme Reaction Mechanisms
Gemma L Holliday, Gail J Bartlett, Peter Murray-Rust, Janet M Thornton, and John BO Mitchell
Unilever Centre for Molecular Informatics, Department of Chemistry, Lensfield Road, Cambridge, CB2 1EW
We have developed an in-house database of enzymatic reaction mechanisms (MACiE). The enzymes used have been selected such that they are well characterised and have crystal structures in the PDB. During the curation of these data it became noticeable that, in this time of increasing knowledge about the structure and mechanism of enzymes, the current classification system has some limitations. Whilst the Enzyme Commission (EC) classification does everything it was ever intended to do, we feel there is a need for an alternative and complementary system which does more.
Here we demonstrate our progress towards a unique and fully searchable database for enzyme reaction mechanisms and other relevant enzymatic data. Our database holds the reaction mechanism for individual enzymes; this includes the overall reaction and the multiple steps that may be involved in the mechanism of action. The data are held in both ISIS/Base and CML (an XML application) formats. Whilst MACiE will continue to be maintained in ISIS/Base in the foreseeable future it is our sincere hope that CML will eventually become the primary format in which MACiE is stored. Preliminary results indicate that MACiE will help us understand the chemistry of enzyme better, allowing us to ask questions ranging from how nature best uses her arsenal of amino acids to how individual amino acids act in any given situation.
References:
P. Murray-Rust, H. S. Rzepa, Chemical Markup, XML, and the World Wide Web. 4. CML Schema, J. Chem. Inf. Comput. Sci., 2003, 43(3), 757-772
Aromatic Hydroxylation by Cytochrome P450 Enzymes
C. M. Bathelt, A. J. Mulholland, J. N. Harvey
School of Chemistry, University of Bristol, Cantock's Close, BS8 1TS
The cytochrome P450 enzymes play a central role in drug metabolism by catalysing the biotransformation of a wide variety of xenobiotics [1]. Understanding the mechanism of these processes is vital for predicting reactivity of chemicals and reducing toxic side effects of drugs.
We have studied the mechanism and selectivity of cytochrome P450-mediated hydroxylation for a number of aromatic compounds using B3LYP density functional theory computations. Our calculations show that addition of the active species Compound I to an aromatic carbon atom proceeds via a transition state with partial radical and cationic character. Ring substituents with both electron-donating and electron-withdrawing properties are found to decrease the addition barrier in para-position. Two new structure-reactivity relationships including these different substituted benzenes have been developed using (i) experimentally derived Hammett σ-constants and (ii) a theoretical scale based on bond dissociation energies of hydroxyl adducts of the substrates, respectively. [2] Furthermore, pathways for oxidation of the drug molecule diclofenac, which contains two aromatic rings, have been explored using a combined quantum mechanical/molecular mechanical (QM/MM) approach. Hereby the steric and polar effects of the enzyme environment of cytochrome P450 2C9 on the hydroxylation of diclofenac were taken into account.
This work provides insight into the detailed mechanism of cytochrome P450 mediated hydroxylation, predicts the electronic contribution of substituents to reactivity and combines intrinsic reactivity with effects of a specific enzyme in the case of diclofenac as an example of human drug metabolism.
References:
1. Anzenbacher, P.; Anzenbacherova, E. Cell. Mol. Life Sci. 2001, 58, 737-747.
2. Bathelt, C. M.; Mulholland, A. J.; Harvey, J. N., J. Am. Chem. Soc. 2003, 125, 15004.
A Molecular Dynamics Study of Vibrationally-Assisted Tunnelling in Enzymes
Linus Johannissen, Prof. Mike Sutcliffe
Adrian Building, University of Leicester, University Road, Leicester, LE1 7RH
The textbook explanation for enzyme catalysis is that the enzyme reduces the height of the energy barrier that the reaction must surmount to move from reactant to product by stabilising the transition state. However, it is becoming apparent that such an over-the-barrier model is not always appropriate: over the past decade or so the concept of quantum tunnelling has been increasingly incorporated into models for enzyme catalysis. The first such models have tunnelling taking place part-way up the energy barrier (the so-called Bell-correction model), but over the past few years it has become apparent that hydrogen transfer reactions can proceed by so-called extreme tunnelling, i.e. tunnelling without having to first partially ascend the barrier. There is experimental evidence that many enzymes catalyse carbon-hydrogen bond cleavage by extreme tunnelling promoted by the vibrations of the protein scaffold - vibrationally-assisted extreme tunnelling. However, different enzyme-catalysed reactions exhibit different temperature-dependency behaviour, leading to the premise that different types of protein vibrations - termed 'active' and 'passive' - can promote tunnelling.
The theoretical understanding of this mechanism is still very much in its infancy. For example, the physical nature of 'active' and 'passive' dynamics, and how they differ, is unknown. Molecular modelling work is being carried out to gain understanding at the atomic level of how vibrational energy is channelled into the active site to promote tunnelling. Molecular dynamics simulations are used to analyse the internal motions of the enzyme, with techniques such as cross-correlation analysis, principal components analysis and Fourier transformations used to identify important changes in the conformation of the enzyme that could promote tunnelling. The energy barrier for the hydrogen transfer is calculated to determine how the internal motions of the enzyme affect the reaction: the shape (height and width) of the barrier through which tunnelling occurs determine the probability of tunnelling, and hence the observed reaction rates. Analysing how vibrations within enzymes affect the energy barrier provides valuable insight into how enzymes promote tunnelling. The results of these modelling studies, and their implications, will be discussed.
References:
Scrutton, N. S. et al. New insights into enzyme catalysis: ground state tunnelling driven by protein dynamics. Eur. J. Biochem. (1999) 264, 666-671
Sutcliffe, M. J. and Scrutton, N. S. A new conceptual framework for enzyme catalysis: hydrogen tunneling coupled to enzyme dynamics in flavoprotein and quinoprotein enzymes. Eur. J. Biochem. (2002), 269, 3096-3102
Multidrug Resistance in Bacteria: Simulation Studies of the AcrAB/TolC System and Related Proteins
Loredana Vaccaro, Prof. Mark S. P. Sansom
Department of Biochemistry, University of Oxford, South Parks Road, Oxford, OX1 3QU
The mostly widely employed mechanism of drug extrusion in bacteria is via membrane transport proteins called efflux pumps. In Gram-negative bacteria, multidrug resistance is conferred by tri-partite complexes, rather than by a single transport protein. Through these systems, a wide range of substrates is expelled from the cytoplasm, through the periplasmic region, to the exterior of the cell. Among these complexes, the AcrAB/TolC system in Escherichia coli is formed by an inner membrane efflux pump, AcrB, an outer membrane protein, TolC, and a periplasmic protein known as an adaptor, AcrA. The components of this complex are studied, in order to provide insights into drug transport in bacteria. Molecular Dynamics simulations in a lipid environment are performed on the transmembrane region of AcrB, whilst the dynamics of MexA, a homologue of AcrA from Pseudomonas aeruginosa, has been studied in bulk water. Both the proteins showed a stable structure during simulations of 20 and 10 ns length, respectively. AcrB displayed a flexible behaviour of the helices thought to be involved in the drug transport from the cytoplasm to the periplasm, TM7 and TM8. A single wire of water molecules interacting with different residues within the pore was observed. This may be used to transfer protons from the periplasm to the cytoplasm, as an H+-transport is the driving force to the drug efflux in AcrB. The dynamics on MexA revealed concerted motions of the two principal domains forming the protein. These can be related to the ability of MexA (and AcrA) to form a complex with the two other components of the complex. Homology modeling of AcrA based on the X-ray structure of MexA has been performed and two slightly different models have been obtained. MD simulations of these two models are being performed to compare their conformational stability.
Distinguishing Structural and Functional Restraints in Evolution in Order to Identify Interaction Sites
Vijayalakshmi Chelliah, Prof. Tom Blundell and Dr. Simon Lovell
Department of Biochemistry, University of Cambridge, 80 Tennis Court Road, Cambridge, CB2 1GA
Structural genomics projects are producing many three-dimensional structures of proteins that have been identified only from their gene sequences. It is therefore important to develop computational methods that will predict sites involved in productive intermolecular interactions that might give clues about functions.
The conservation of amino acid residues has been shown to be strongly dependent on the environment in which they occur in the folded protein and amino acid substitution tables that give the likely substitutions of amino acids in particular local environments have been derived [1,2]. A method to distinguish those restraints placed on protein structure from additional restraints due to particular functions mediated by interactions with other molecules has been developed using these environment-specific substitution tables (ESSTs) [3]. The positions where environment-specific substitution tables make poor predictions of the overall amino acid substitution pattern are identified using information theory calculations [4]. The scores derived from these methods are mapped onto the protein three-dimensional structures and contoured, allowing identification of clusters of residues with strong evolutionary restraints. We find that the clusters of high scoring alignment positions apparently subjected to these additional restraints in evolution correlate well with the functional sites in protein defined by experimental methods. The method is applied to a set of well-characterised protein families and is able to identify functional sites. The technique is fast, automatic and predicts functional sites with a high degree of accuracy. Finally, we examined the use of functional site prediction in protein-protein docking.
References:
1. Overington, J., Donnelly, D., Johnson, M. s., Sali, A. & Blundell, T. L. (1992) Protein Science, 1, 216-226.
2. Overington, J., Johnson, M. S., Sali, A. & Blundell, T. L. (1990) Proc. Roy. Soc. London ser.B, 241, 132-145.
3. Chelliah, V., Chen, L., Blundell, T. L. & Lovell, S. C. (2004) J Mol Biol 342: 1487-1504.
4. Yona, G. and Levitt, M (2002) J Mol Biol 315: 1257-1275.
Novel Computational Methodology for analysing protein ligand/binding sites by use of Vicinity Analysis
Arthur Mc Gready (1), Brian Hudson (1) and David Whitley (1), Adrian Stevens (2) and Mike Lipkin (2)
1. Centre for Molecular Design, Institute of Biomedical and Biomolecular Science, University of Portsmouth, King Henry I Street, Portsmouth PO1 2DY
2. Biofocus, Chesterford Research Park, Saffron Walden, Essex, CB10 1XL
A novel method for analysing protein binding sites will be described. The method takes modified PDB files that contain only contact residues and a drug fragment and compares them against a specific target (defined set of residues or other binding site). A set of ranked matches is generated for further analysis.
From an evolutionary viewpoint all protein superfamilies share sets of specific characteristics. These include residues involved in the binding sites of particular proteins. Traditional methods for analysing these characteristics include phylogenetic analysis (BLAST), but these methods do not always produce reasonable results. Vicinity Analysis is a novel 3D method for analysis of these structures. This involves generating an adjacency matrix from the following criteria: Chemical Similarity [1] and Intra Molecular Distances. The adjacency matrix is then processed by means of Clique Analysis [2] to find any sets of similarities between contact residues and the user defined target file.
A data set of 129 protein structures was selected for analysis. These include a range of different superfamilies including Kinases, Transferases, Proteases and Hydrolases. These files were processed by use of software to remove all irrelevant data leaving only contact residues and the drug fragment. The data set is then analysed using Vicinity Analysis and all results are ranked using the RMSD for each particular protein binding site (sub-pocket) target and user defined query. Other factors taken into account include the maximal clique size generated by the clique analysis. With each clique a set of α-carbon to β-carbon angles are generated to give the orientation of the amino acid side chain residues in relation to the ligand-binding site.
This method is an alternative to traditional in silico docking methods to help in the development of drug design. Traditional docking methods can produce adequate docking poses but cannot produce a sutable ranking system to produce the best fit. The Vicinity Analysis method employs the use of experimental data to identify small molecule fragments likely to bind into a binding site (sub-pocket) and so recombination of the cliques could produce new drug molecules.
Results to date have shown that Vicinity Analysis produce results consistent with other similar studies [3] based only on the α-carbon distances. The inclusion of other tolerances, for example the directionality of the α-β bond vectors and semi-dynamic parameters such as the crystallographic B-factors, improves the representation.
References:
1. S.Helberg, M Sjostrom, B Skagerberg and Swante Wold. Peptide Quantitative Structure Relationship, A Multivariate Approach. J. Med Chem (1987), 30 (1126-1135).
2. Bron C., Kerbosch J., Algorithm 457. H. CACM (1973), 16 (575-577).
3. Gardiner E.J. and Willet P. Journal of Chemical Information Computer Science (2000), 40 (273-279)
Developing a protein - ligand docking algorithm: Flexligdock
P.R. Oledzki and R.M. Jackson
School of Biochemistry and Microbiology, University of Leeds, Leeds, LS2 9JT, UK
A program Flexligdock is being developed to provide a flexible ligand docking tool for small molecule docking to proteins. The program is based on Q-fit [1] a rigid protein-ligand docking algorithm and uses a probabilistic sampling method in conjunction with the GRID [2] molecular mechanics force field to generate and score solutions.
The method fragments a ligand at each rotatable bond to produce a series of rigid fragments termed seed fragments (or anchors). Flexligdock then utilises an interaction point methodology to map the ligand fragments onto interaction energy grid maps of the protein target. It then applies an incremental construction algorithm to build the ligand, fragment by fragment, in the protein binding site. This stage employs torsion angle sampling to permit simulation of ligand flexibility. The algorithm has been parameterised on a data set of 46 protein-ligand complexes. The parameterisation data set contains a structurally diverse set of proteins and a variety of ligands that contained between 0-23 torsion angles.
Three main parameters used for filtering out poor solutions during incremental construction were investigated. The filtering parameters investigated were; (1) the intra-molecular ligand collision distance, (2) the percentage threshold for inter-molecular ligand-receptor collisions and (3) the number of solutions passed to each stage of the incremental construction algorithm. With an optimal parameter setting Flexligdock docked 46 out of the 55 ligands <2Å RMSD as the top ranked solution with the manual seed placement. However, with automatic seed placement only 25 out of the 55 ligands were docked <2Å RMSD as the top ranked solution. This result proves that the incremental construction algorithm produces good ligand conformations with correct seed positioning.
The FlexX validation data set of 200 protein-ligand complexes [3] has been docked with Flexligdock to permit comparison against other existing protein-ligand docking algorithms. The FlexX docking algorithm docks 46% of the data set as the top ranked solution. Flexligdock (with automatic seed placement) docked 47% of the data set as the top ranked solution. Currently, further investigation of parameters and improvements to the docking algorithm are being undertaken to increase the accuracy of Flexligdock.
References:
1. Jackson, R.M. (2002). Q-fit: a probabilistic method for docking molecular fragments by sampling low energy conformational space. J Comput Aided Mol Des. 16(1):43-57
2. Goodford, P.J. (1985). A computational procedure for determining energetically favourable binding sites on biologically important macromolecules. J. Med. Chem 28, 849-857
3. Kramer B. et al (1999). Evaluation of the FlexX incremental construction algorithm for protein-ligand docking. Proteins. 37:228-241
Identification of novel anticancer agents with 'forward-reverse' virtual high throughput screening
Andrew Knox (1), Dermot Frost (2), Mary Meegan (1), David Lloyd (3).
1. Department of Pharmaceutical Chemistry, School of Pharmacy, Trinity College Dublin
2. Trinity Centre for High Performance Computing
3. Department of Biochemistry, Trinity College Dublin
Approximately 42,000 cases of breast cancer are diagnosed annually in the United Kingdom and Ireland. 14,000 of these cases are terminal.
In breast cancer, 60% of Estrogen Receptor (ER) positive tumours utilize ER alpha to activate growth regulatory genes and stimulate proliferation. Our work focuses on ER alpha as a therapeutic anticancer target, with the expectation that inhibitors can prevent the growth, spread and recurrence of breast cancer.
We present the discovery of potent inhibitors of ER alpha from our virtual high throughput screening protocol involving a computational platform optimised and validated for breast cancer discovery research, using x-ray crystal models of ER alpha in docking simulations.
We provide an assessment of compound database pre-processing and explore the efficacy of a variety of commercial virtual screening platforms applied to the estrogen receptor.
Employing a 160 processor cluster, our 'in house' system is capable of screening up to 50,000 compounds per day considering both ligand and receptor flexibility. A two-tiered scoring scheme is adopted, to which optimal scoring geometries obtained from the docking procedure are retained and re-scored by an optimal scoring function.
Virtual hits identified are subjected to clustering analyses, and subsets chosen for testing using a biochemical assay. Inhibitors identified from our protocol have been confirmed by biological assay, and subsequently computationally back-docked through proteins involved in key cellular processes that may result in toxicity or side effect, to establish other possible interactions and better profile resulting lead compounds.
Multiscale Docking Using Evolutionary Optimisation
David J. Huggins, Guy H. Grant
Department of Chemistry, University of Oxford, Central Chemistry Laboratory, South Parks Road, Oxford, OX1 3QH, UK
OXDOCK [1] is a grid-based multiscale docking algorithm that calculates the van der Waals and electrostatic interactions between a protein and a chosen ligand [2]. Recent validations have shown that OXDOCK is a useful tool for finding binding sites and exploring docking modes [3]. Development of OXDOCK has led to the creation of Eve, a docking algorithm that combines the speed of a multiscale approach with the optimizing ability of evolutionary programming. Initial docking results are calculated using single, double and triple point representations of the ligand to fix the geometry of the ligand. The best docking modes are then used as the first generation of a genetic algorithm which explores the position and conformation of the ligand. Extra energy terms have been added to simulate the hydrophobic effect and the effect of rotor restriction. Initial results, on a test set of over 100 protein-ligand complexes from the Protein Data Bank, show that Eve can simulate molecular docking in a small amount of computer time. Further modifications suggest that Eve can be a useful tool for lead optimization using evolutionary programming, by the successive addition of molecular fragments.
References:
1. Glick, M.; Grant, G. H.; Richards, W. G. Docking of flexible molecules using multiscale ligand representations. Journal of Medicinal Chemistry 2002, 45, 4639-4646.
2 Glick, M.; Grant, G. H.; Richards, W. G. Pinpointing anthrax-toxin inhibitors. Nature Biotechnology 2002, 20, 118-119.
3. Dubos, C.; Huggins, D.; Grant, G. H.; Knight, M. R.; Campbell, M. M. A role for glycine in the gating of plant NMDA-like receptors. Plant Journal 2003, 35, 800-810.
The Use of Free Energy Simulations as Scoring Functions
Julien Michel, Richard D. Taylor and Jonathan W. Essex
Department of Chemistry, University of Southampton, Highfield, Southampton, SO17 1BJ
The accurate prediction of free energies of binding is of great importance for the computational design of drug candidates. Although they are very fast, traditional scoring functions are not sufficiently accurate for the computational optimisation of a hit, and are usually limited to picking promising candidates from large databases of compounds. However, free energy simulations, whose principles are grounded in statistical mechanics, have been used for two decades and can achieve a significantly higher accuracy in the prediction of binding free energies. Unfortunately, the computational expense associated with this method prohibits its application to large number of compounds.
In this project, we seek to develop a protocol based on free energy simulations that is computationally efficient and yet still retains the accuracy of conventional free energy methods. To achieve this aim we have parameterised a continuum model of water compatible with the AMBER forcefield and based on the Generalised Born Surface Area (GBSA) formalism. (1,2) The Pairwise Descreening Approximation (PDA) method (3) has been adopted to compute the Born radii required by the GBSA theory. Different parameterisation protocols have been investigated. The implicit models of water have been developed by fitting adjustable parameters to experimental hydration free energies of a dataset of small molecules that covers a wide range of chemical functionalities. The parameterised models have been tested by computing potentials of mean force for the association of small molecules that are typical of hydrophobic, aromatic, polar and ionic interactions, and representative of amino-acid side chains. Our results highlight weaknesses in a previously used parameterisation protocol. The best water model has been implemented in an in-house Monte Carlo engine and a robust protocol that allows for the fast calculation of the difference in the free energy of binding of structurally diverse ligands is now being developed. Preliminary results will be reported.
References:
1. Michel J., Taylor R. D. and Essex J. W., J. Comp. Chem. (2004) 25, 1760
2. Still W. C. et al., J. Am. Chem. Soc. (1990) 112, 6127
3. Hawkins G. D. et al., Chem. Phys. Lett. (1995), 246, 122
Generation of Multiple Pharmacophore Hypotheses Using a Multiobjective Genetic Algorithm
Simon J. Cottrell (1), Valerie J. Gillet (1) and Robin Taylor (2)
(1) Department of Information Studies, University of Sheffield, Western Bank, Sheffield, S10 2TN, UK
(2) Cambridge Crystallographic Data Centre, 12 Union Road, Cambridge, CB2 1EZ, UK
Pharmacophore generation involves overlaying a set of ligands such that functional groups relevant to biological activity are overlaid in the most plausible way. Existing methods for pharmacophore generation generally take one of two approaches to handling conformational flexibility of the ligands. Firstly, there are methods which essentially use rigid-body techniques, but process several pre-generated conformers of each input molecule sequentially. Secondly, there are methods that search the conformational space of the molecules dynamically during the overlay process, so as to find the set of conformations that produce the best overlay. The former generally return a large number of solutions which take considerable effort to analyse and are highly dependent on the method used to generate to the conformers, whilst the latter generally return a single solution, implying an unrealistic degree of certainty in the result.
This work has involved applying a multiobjective genetic algorithm (MOGA) to the pharmacophore generation problem [1]. The MOGA generates several solutions but takes the second of the above approaches to conformational flexibility. Three criteria, or objectives, are considered in evaluating hypotheses, namely the closeness of the alignment of features in the different ligands, the volume overlap of the ligands and the internal energy of the ligands. In a conventional approach these would generally be combined into a single function with arbitrary weighting, and a single solution that maximises this function would be returned. In the multiobjective approach, however, they are not combined into a single function. Instead, the program generates several different hypotheses which represent different, but equally valid compromises between the objectives, according to the principles of Pareto dominance.
An important aim of this work has been to generate a set of solutions that are diverse from a biochemical point of view. Ensuring a diverse range of different compromises between the three objectives has proved to be a necessary but not sufficient condition for achieving this. Considerable effort has therefore been directed towards explicitly taking into account chemical diversity within the MOGA population.
The results of the MOGA will be illustrated using datasets for three binding sites of pharmaceutical interest. In each case, the MOGA generates a manageable number of different hypotheses. Thus, it takes a realistic view of the uncertainty which is inherent when the binding site structure is not known, but limits its output to those hypotheses that are plausible based on previously established criteria.
References:
1. Cottrell, S.J.; Gillet, V.J.; Taylor, R.; Wilton, D.J. Generation of Multiple Pharmacophore Hypotheses. Journal of Computer-Aided Molecular Design, in press.
