2 Physical Methods and Techniques Part (ii) Computer Graphics (Computer Aids to Organic Chemistry) By C. I. DE MATTEIS D. E. JACKSON and N. RAJ Department of Pharmaceutical Sciences University of Nottingham Nottingham NG7 ZRD UK 1 Introduction This is the third in a series of Reports describing the use of computer graphics in organic chemistry and follows on from the previous Reports in 1988 and 1990. As one might expect the last two years have witnessed an increased use of computer graphics in a number of applications including ‘windowed’ interfaces both to a variety of modelling software and also as a powerful visualization tool within the research environment. Graphically orientated applications such as database and data retrieval systems expert chemical systems and molecular modelling software responsible for calculating and displaying molecular structure and properties have all developed significantly over this period.In addition the potential of the computer graphics image as a means of displaying large quantities of data and in helping understand complex three dimensional problems is increasingly recognized.’ Integration of these various applications into a single graphical system allowing ease of use and greater efficiency in data management continues and a number of commercial software companies now provide such open interfaces. Additionally newly developed computer based tech- niques whereby colour graphics video animation and sound are integrated into a single computer display are being widely exploited in computer aided learning thus introducing computer graphics into the teaching environment.2 Increasingly easy access to powerful computing facilities at ever declining prices has fuelled the observed growth in computer graphics applications.It is estimated that CPU speed is doubling every 18 months,1c1 and that the ratio of price to performance is decreasing by an order of magnitude every 5-7 years3 Also graphical displays will continue to improve in clarity and realism as hardware power develops. Additionally the increasing speed of network hardware will allow graphics images and animations to be sent very rapidly to locations around the world. SuperJANET (Joint Academic Network) which is being installed in UK universities in 1993 will initially carry 140 ((I) A.J. Olson and D. S. Goodsell Scient$c American. Nov. 1992 44:(h)J. Petts. Lob. Pructicr. 1991. 40(7).9. * ‘Multimedia The CTISS File’. ed. J. Darby. CTISS Publications. Oxford. 1992. W. F. van Gunsteren and H. J.C. Berendsen. Angeat. Chem.. Int. Ed. Engl. 1992 29 992. 21 C.I. de Matteis D. E. Jackson and N. Raj million bits of information per second compared to 2 million bits per second delivered by the original JANET.4 The last two years have consequently seen an increase in the number of specialized research journals describing the use of computer aids in chemical and biological research reflecting increasing recognition of the value of these research tools. The Journal of Computer- Aided Materials Design,’ Molecular Modelling and Computational Chemistry Results,6 and Perspectives in Drug Discovery and Design’ are new journals in this field.Protein Science’ is published with a new interactive graphics supplement on diskette which allows the display and manipulation of graphical data. A number of text books have been published in this field’ and certain texts e.g. ‘Reviews in Computational Chemistry’ are now produced annually.’’ This Report will attempt to provide an update to the previous Annual Report of 1990. Given the enormous breadth of the field no attempt will be made to provide a thorough review of the literature but rather to provide an insight into the important developments in computer graphics applications over the last two years.2 Molecular Modelling Drug Design.-It is estimated that in 1989 US pharmaceutical companies spent $5.71 billion on domestic Research and Development to produce a total of 23 new US approved drugs.” The need to increase the role of rational drug design in the drug discovery process is obvious and molecular modelling is used increasingly within this process. Examples of the successful use of molecular modelling in drug design have been published,I2 and there are an increasing number of examples of its use in the de now design of protein^.'^ A number of articles and books reviewing the methods and applications of molecular modelling in the design process have a~peared,’~ together with those describing the use of these techniques to improve our understanding of biological events.’” Molecular modelling approaches to rational drug design utilize available informa- tion about the interaction of active substrate molecules with the biological macro- molecule target site.Depending on whether the three dimensional structure of the biological target molecule is available or not two distinct approaches to the design E. Geake New Scientist 1992 1848 18. Journal of Computer-Aided Materials Design ESCOM Science Publishers Leiden The Netherlands. Molecular Modelling and Computational Chemistry Results MMCC Publishing Massachusetts USA. ’ Perspectives in Drug Discovery and Design ESCOM Science Publishers Leiden The Netherlands. Protein Science Cambridge University Press Cambridge UK. (a) ‘Computer Simulation of Biomolecular Systems’ ed.W. F. van Gunsteren and A. J. Wilkinson ESCOM Leiden 1993 vol. 2; (h) ‘MOTECC-Modern Techniques in Computational Chemistry’ ed. E. Clement] ESCOM Leiden 1991. ‘Reviews in Computational Chemistry’ ed. D. B. Boyd and K. B. Lipkowitz VCH Publishers Weinheim 1992 vol. 3. M.E. Wolff and E.T. Maggio Med. Chem. Res. 1991 1 101. (a)A. Olson and D. Goodsell Curr. Opin. Struct. Biol. 1992,2 193; (b)J. Hodgson BiolTechnoloqy 1991 9 19. l3 A. Pessi E. Bianchi A. Crameri. S. Venturini A. Tramontano and M. Sollazzo Nature (London) 1992 362 367. I4 (a) W.C. Ripka and J.M. Blaney Top. Stereochem. 1991 20 1; (b) Y:C. Martin in ‘Methods in Enzymology’ ed. J. N. Abelson and M. I. Simon Academic Press 1991 vol. 203 p. 587; (c)J. S. Dixon Trends Biotech.1992 10,357; (d)J. P. Snyder Med. Res. Rev. 1991 11(6) 641; (e)‘A Textbook of Drug Design and Development’ ed. P. Krogsgaard-Larsen and H. Bundgaard Harwood 1991. Physical Methods and Techniques -Part (ii) Computer Graphics process are available. These have been referred to respectively as 'direct' or 'indirect' ligand design. In those cases where the three dimensional structure of the target site is known the drug design process involves three components. Initially the position of the binding site must be ascertained followed by an understanding of the mode of association of the ligand within this target site and finally using this information either de nouo ligand design or lead optimization is attempted. Examples of the use of NMR15 and X-ray crystallography' to study macromolecule-ligand complexes have appeared and it is claimed that once the structure of a biological macromolecule has been solved crystallographically the structure of the complex can be ascertained in a number of days.' 6*14aExamples of multiple ligand binding modes have been observed experimen- tally.' 4* The incorporation of crystallographic studies of ligand binding into enzyme inhibitor design strategies has been reported resulting in four structurally distinct novel inhibitors of Escherichia coli thymidylate synthase.Rational de nouo drug design methods generally involve the analysis of the target binding site and based on this information the prediction of compounds which may bind. A number of computational methods have been developed that locate favoured positions for ligand functional groups within the binding site.For example GRID locates the positions where particular ligand atoms or functional groups will prefer to bind by calculating the interaction energy of this functional group with the receptor at various positions around the active site using a potential energy function which considers van der Waals electrostatic and hydrogen bonding interactions.' This method has recently been upgraded to consider more accurately the possibility of the ligand functional group forming more than one hydrogen bond to the active site.'' HSITE produces a map of hydrogen bonding regions within an enzyme active site." A method for determining and displaying the position of hydrophobic and hydrogen bonding areas in ligand binding sites has been described.20 Once favourable positions for these ligand fragments have been located three dimensional databases can be screened for molecules that match these pharmacophore requirements.2 ' Alternatively fragments may be pieced together with a trial-and-error approach using either molecular graphics facilities or techniques that have been developed to assemble fragments in the required orientation for biological activity.A number of database methods have been developed that search for substructures which can act as templates and hold functional groups in the required orientations.2' CAVEAT uses a database of cyclic compounds that can be used as 'spacers' to connect the fragments required for ligand activity.22 LUDI calculates the position of interaction sites where l5 S.W.Fesik J. Med. Chem. 1991 34 2937. l6 K. Appelt R. J. Bacquet C.A. Bartlett C. L. J. Booth S.T. Freer M. A. M. Fuhry M. R. Gehring S. M. Herrmann E. F. Howland C.A. Janson T. R. Jones C.-C. Kan V. Kathardekar K. K. Lewis G.P. Marzoni D.A. Matthews,C. Mohr E. W. Moomaw,C. A. Morse,S. J.Oatley R. C. Ogden M. R. Reddy S.H. Reich W. S.Schoettlin W. W. Smith M.D. Varney J. E. Villafranca R. W. Ward S. Webber S.E. Webber K.M. Welsh and J. White J. Med. Chem. 1991.34 1925. l7 (a) P. J. Goodford J. Med. Chem. 1985,28 849; (h)D. J. Boobbyer P. J. Goodford P. M. McWhinnie and R.C. Wade J. Med. Chem. 1989 32 1083. l8 R.C. Wade K.J. Clark and P.J. Goodford J. Med. Chem.1993 36 140. l9 (a)D. J. Danziger and P. M. Dean Proc. R. SOC.London Ser. B. Biol.Sci. 1989,236,101;(b) ibid. 1989,236 115. 'O R.S. Bohacek and C. McMartin J. Med. Chem. 1992 35 1671. " Y.C. Martin J. Med. Chem. 1992 35 2145. " 'CAVEAT' G. Lauri G. T. Shea S. Waterman S.J. Telfer,and P. A. Bartlett University of California at Berkeley USA. 24 C.I. de Matteis D. E. Jackson and N. Raj hydrogen bonding and hydrophobic fragments should be placed using rules derived from an analysis of non-bonded contacts found in the Cambridge Crystallographic Database and will also use the output from GRID calculations. Up to four of these resulting interaction sites are then built into fragments using a library of approximately 600 ‘linkers’ and these fragments are then joined up into the complete molecule using small ‘bridging’ fragments which include methylene and carboxylate.This method has been used to generate improved inhibitors for dihydrofolate reductase and HIV pr~tease.’~ A structure generation algorithm that will build structurally diverse three dimen- sional chains joining fragments of defined positions within a constricted site for example an enzyme active site has been de~cribed.’~ BUILDER is a modelling program that combines automatic structure generation with database searching approaches to generate lead molecules in which the required fragments are joined.25 Chau and Dean have reported the generation of a database of small organic fragments which may be used in the future for automated site-directed drug design.26 Alternative methods for designing ligands to fit a known active site include the following DOCK which carries out a steric search of three dimensional databases to locate compounds that are geometrically and electrostatically complementary to the binding site;” LEGEND which attempts to grow a ligand within the binding site by building up the structure sequentially from random atom types and dihedral angles;28 and GROW which aims to determine the optimal peptide ligand for a given enzyme by gradually building the ligand using a library of amino acid conformation^.^^ Interactive molecular modelling approaches involving the docking of novel ligands into the target site using distance constraints to position the ligand have been rep~rted.~’ Free Energy Perturbation methods are now well established for calculating the difference in free energy of the binding of related ligands to a given target ~ite.~~v~~~ In those cases where the three dimensional structure of the target molecule is not known a variety of approaches are available for inferring the structure of the target site.If the protein of interest has homology with a protein of known three dimensional structure modelling techniques have been developed that will generate a three dimensional structure for this protein,31 although it is recognized that this model is less reliable than an experimentally determined str~ture.~~~ The use of the binding site from an enzyme with the same mechanism of action has been re~0rted.l~~ Antibody technology is being used in a variety of ways to provide information about the target site structure.By raising antibodies to the substrate or drug molecule a receptor mimic is produced and structural analysis of the relevant part of the antibody will provide information about the receptor structure.32 Another approach has been to produce an 23 (a)H. J. Bohm J. Cornput.-Aided Mol. Design 1992 6 61; (b) ibid. 1992 6 593. 24 R.A. Lewis J. Mol. Graph. 1992 10 131. 25 R. A. Lewis D. C. Roe C.Huang T. E. Ferrin R. Langridge and I. D. Kuntz J. Mol. Graph. 1992,10,66. 26 (a)P.-L. Chau and P. M. Dean J. Cornput.-Aided Mol. Design 1992,6,385; (b)ibid. 1992,6,397; (c)ibid. 1992 6 401. 27 R. L. DesJarlais R. P. Sheridan. G. L. Seibel J. S. Kuntz and R.Venkataraghavan J. Med. Chem. 1988 31 122. Y. Nishibata and A. Itai Tetrahedron 1991 47 8985. 29 J.B. Moon and W.J. Howe Proteins Struct. Funct. Genet. 1991 11 314. 30 T.J. Mitchell J. Mol. Graph. 1992 10 53. 31 S. Kawakita R. Kuroki. and T. Yao J. Mol. Graph. 1992 10 58. 32 M. A. Sherman and M.B. Bolger J. Bid. Chem. 1988 263 4064. Physical Methods and Techniques -Part (ii) Computer Graphics 25 antibody to the target site which will produce a mirror image of the target site structure. This antibody can then be analysed structurally to obtain pharmacophoric information or used as an antigen to another antibody so that structural analysis of this antibody will provide information about the target site structure." AbM is a program allowing antibody three-dimensional structure modelling from sequence information using a knowledge base of structural information combined with computational methods.33 Receptor mapping represents another approach for inferring the structure of the target site but requires no information about the location or structure of the target macromolecule By systematically altering the structure of the lead compound and ascertaining its activity the functional groups or atoms required for binding or activity can be identified and the three dimensional orientation of these molecular features provides a pharmacophore or receptor map.Deciding how to superimpose the individual molecules so as to align the pharmacophoric groups represents a considerable problem particularly if the molecules are structurally diverse and have considerable conformational flexibility.14' Both an automatic method for positioning two molecules so that the similarity between their molecular electrostatic potentials is at a maximum34 and molecular matching using simulated annealing have been de~cribed.~' SUPER matches the van der Waals surfaces and charge distributions of two molecules to produce optimum overlap whilst OVID finds the best surface overlap of user-defined atoms in two rnolec~les.~~ A method which superimposes structures on the basis of similarity of molecular shape hydrogen bonds and electrostatic interactions and additionally considers the conformational flexibility of the molecules has been rep~rted.~ Molecular similarity calculations have also been used as a method for screening large structural data sets obtained from three dimensional databases and has been shown to provide a rapid screening method.38 Quantitative structure activity relationships (QSAR) where biological activity in a series of congeneric compounds is described by a series of molecular descriptors continue to be used and developed for drug design application^.^^ Novel parameters continue to be developed for example parameters derived from theoretical chemistry4' and a novel hydrophobic parameter4' have been reported.Neural networks have been found to produce superior statistical results in QSAR when compared to regression analysis.42 Force Fields.-The motivation for molecular modelling is to try to predict physical properties of molecular systems based on the interactions that determine their behaviour.This is achieved at the most fundamental level by invoking quantum mechanics through use of the Schrodinger equation. Such ab initio or first principle 33 'AbM' Oxford Molecular Limited The Magdalen Centre Oxford Science Park Sandford-on-Thames Oxford OX4 4GA UK. 34 F. Manaut F. Sanz J. Jose and M. Nilesi J. Cornput.-Aided Mol. Design 1991 5 371. 35 (a)M. T. Barakat and P.M. Dean J. Cornput.-Aided Mol. Design 1990,4 295; (b) ibid. 1990,4 317; (c) ibid. 1991,5 107; (d)M. C. Papadopoulos and P. M. Dean J. Cornput.-Aided Mol. Design 1991 5 119. 36 R. B. Hermann and D. K. Herron J. Cornput.-Aided Mol. Design 1991 5 51 1. 37 Y. Kato A. Inoue M.Yamada N. Tomioka and A. Itai J. Cornput.-Aided Mol. Design 1992 6 475. '13 A.C. Good E.E. Hodgkin and W.G. Richards J. Cornput.-Aided Mol. Design 1992 6 513. 39 E.J. Ariens Quant. Struct.-Act. Relat. 1992 11 190. 40 L.Y. Wilson J. Med. Chern. 1991 34 1668. 41 Y.-Z. Da K. Ito and H. Fujiwara J. Med. Chern. 1992 35 3382. 42 S.4. So and W.G. Richards J. Med. Chern. 1992,35 3201. 26 C.I. de Matteis D. E. Jackson and N. Raj calculations rapidly become unfeasible however as the atomic system size starts growing. In contrast to the more theoretically based Molecular Orbital methods classical approaches such as Metropolis Monte Car10,~~ Molecular Mechanics (or energy minirni~ation),~~ Molecular dynamic^,^' and Distance Geometry46 utilize empirical or semi-empirical potential-energy functions.Such a function describes intramolecular forces and consists of terms to include bonded and non-bonded interactions. For the bonded interactions there are two-body bond length vibrations three-body angle bending vibrations and four-body dihedral rotations. The non-bonded terms usually incorporated into the force field include electrostatic and van der Waals interactions. Hydrogen-bonding may or may not be explicitly included. In cases where it is not appropriate adjustments to atomic charges may be made. An energy function of this type is easily differentiable and so allows forces on each particle to be evaluated as a function of position coordinates. The total Hamiltonian of the system is then given by the sum of various potential energy contributions plus the kinetic energy of each constituent.Equation 1 shows a typical potential energy form. V,, = Z bond stretching + E angle bending + E dihedral rotation + C (VDW + electrostatic) A wide variety of functional forms for each of the terms in equation 1have been used. In the case of bond stretching the most common is the Hooke’s law harmonic potential. This form is however only valid for small deviations of the bond from its ‘reference’ length and away from this alternatives such as the Morse47 potential are more appropriate. For angle bending again harmonic potentials are most commonly used but for greater accuracy higher order terms in the Taylor expansion for the potential are required. Dihedral angles exhibit multiple minima and it is necessary to use a periodic cosine function to describe these potentials.This usually takes the form of a truncated Fourier series involving one- two- and three-fold terms. For non-bonded atomic pairs the majority of current molecular mechanics force fields use a combina- tion of van der Waals and Coulombic energy terms. The other factor determining the success of a force field is the parameter set chosen to be used within the potential functional form. Such parameters are required for equilibrium bond lengths and bond angles with their respective force constants; dihedral barrier heights and phases; out-of-plane terms; van der Waals radii and energy well depths; partial atomic charges or bond dipoles; and perhaps a variety of cross-linking terms.Further these parameters must be chosen for all atom types comprising the class of molecules to be studied. It should always be possible to reproduce experimental results to a high degree of accuracy given arbitrarily complex analytical potential functions and highly specific atom types; however such a force field is probably only of limited use. Another approach is to sacrifice accuracy for ” N. Metropolis A. W. Rosenbiuth M. N. Rosenbluth and A.H. Teller 1.Chem. Phys. 1953 21 1087. 44 J. A. McCammon and S.C. Harvey ‘Dynamics of Proteins and Nucleic Acids’ Cambridge University Press Cambridge 1987. 45 M. P. Allen and D. J. Tildesley ‘Computer Simulation of Liquids’ Clarendon Oxford 1987. 46 G. M. Crippen and T. F. Havel ‘Distance Geometry and Molecular Conformation’ Research Studies Press Taunton 1988.47 S.D. Morley R.J. Abraham I.S. Haworth D.E. Jackson M.R. Saunders and J.G. Vinter J. Cornput.-Aided Mol. Design 1991 5 475. Physical Methods and Techniques -Part (ii) Computer Graphics 27 greater transferability and wider range of application. Below we describe recent advances in force field development together with their range of application. In addition we also review some articles dealing with the important topic of electrostatics. The standard force fields are limited to particular combinations of atoms and have been devised to be used in modelling specific molecular systems. Rappe and co-workers have recently reported on a new force-field called the Universal force field (UFF),4s in which the functional forms parameters and generating formulae are intended for the full periodic table.The set of fundamental parameters within UFF is based on the element its hybridization and its connectivity; in total UFF includes 126 atom types. Detailed comparisons are made of conformational energetics and molecular structures with experimental and published MM2 and MM3 results for organic molecules.49 The ability of UFF to reproduce the structures of a variety of main group molecules is also examined.” Whilst UFF represents a new force field the long-standing ones continue to be updated and refined. One such case is the study of the MM2 force field concerning electrostatic corrections proposed by Allinger and Lii,’ and independently by Pettersson and Liljefors.s2 Modelling was performed on a set of compounds containing phenyl groups and polar substituents on neighbouring carbon atoms with the inclusion of dipole moments for Csp2-Csp3 and Csp2-H bonds.s3 The resulting conformations with these modifications show significant improvements though a lower value for the V parameter for the torsional unit C,p2-Csp3-Csp3-Csp2 was used instead of the MM2 value.The potential functions for simple amides several peptides and the small protein Crambin was recently reported for the MM3 force field.s4 The force field for simple amides gives good results for either gas-phase or crystal structures and fair results for vibrational spectra. For peptides and the protein Crambin structural results are comparable to more specialized protein force fields.Results of calculations on aldehydes and ketones using MM3 have also recently been reported.” An alternative to the all-atom modelling of peptides and proteins is the approach taken by Gerbe~-’~ in which entire residues are the smallest units. In this description each amino acid is represented by a single point in space taken to be the position of the a-carbon atom. Additional degrees of freedom are the torsional angles + and t,b to account for the orientation of the peptides links. This ‘peptide mechanics’ force field is reported to reproduce secondary structure elements with high accuracy. Hoops et a1.57 describe application of a systematic method for incorporating a metal ion and its ligand into a classical force field.In particular reference 57 extends the AMBER force field to model the Zinc ion in human carbonic anhydrase I1 in both high and low pH forms. The approach should be transferable to computational studies of other metalloproteins at fixed coordination numbers. 48 A. K. Rappe C. J. Casewit K. S. Colwell W. A. Goddard and W. M. Skiff J. Am. Chem. SOC. 1992,114 10024. 49 C.J. Casewit K.S. Colwell and A.K. Rappe J. Am. Chem. Soc. 1992 114 10035. ’O C. J. Casewit K. S. Colwell and A. K. Rappe J. Am. Chem. Soc. 1992 114 10046. ” N. L. Allinger and J. H. Lii J. Comput. Chem. 1987 8 1146. ’* I. Pettersson and T. Liljefors J. Comput. Chem. 1987 8 1139. ’3 P. M. Ivanov and T. G. Momchilova J. Mol. Strucr. (Theochern) 1991 233 115.54 J. H. Lii and N. L. Allinger J. Comput. Chem. 1991 12 186. ” N. L. Allinger K. Chen M. Rahman and A. Pathiaseril J. Am. Chem. SOC. 1991 113 4505. ’6 P. R. Gerber Biopolymers 1992 32 1003. ” S.C. Hoops K. W. Anderson and K.M. Merz J. Am. Chem. Soc. 1991 113 8262. 28 C.I. de Matteis D. E. Jackson and N. Raj The COSMIC force field developed by Vinter and co-workers over the past fifteen years has primarily been used for modelling relatively small organic molecules. Over this period of software development the potential functions parameters and atom types have been kept as simple as possible with neither cross-terms nor higher-order terms to describe bond or angle distortions. Atom types describe little more than the hybridization and basic geometry of each atom.Morley et have recently reported a number of modifications to COSMIC. These include a two-parameter Morse potential in place of the Hill potential to describe non-bonded interactions; the introduction of a simple iterative Huckel n-electron molecular orbital calculation to allow modelling of conjugated systems; the use of explicit hydrogen-bonding potentials; and the introduction of new atom types. First generation force fields were restricted in their complexity of the analytic representation of the energy surface due to the limited amount of experimental data available for parameterization. The solution to the sparseness of such data came through the advance of high-quality ab initio generated potential-energy surfaces.The approach is to perform ab initio calculations on a number of distorted structures of a given system and so generate large amounts of data describing the full anharmonic potential-energy surface. The aim of Class I1 force fields then is to reproduce structures energies vibrational frequencies and other observables to a high degree of accuracy using a single set of parameters even for problem systems such as highly strained rings. Current Class I1 force fields include Allinger’s MM3’* and Biosym’s CFF91.” The CFF91 force field which now allows modelling of amides and amines,60 employs a quartic polynominal for bond stretching and angle bending; torsions are described by a three-term Fourier expansion and an out-of-plane term is also included. There are seven off-diagonal terms together with a Coulombic interaction between atomic charges.Finally a 9-6 potential is used for van der Waals interactions. Determination of forcefield parameters has been done by Aleman et ~1.~’ by use of quantum mechanic calculations. The approach has been incorporated into a computer program called PAPQMD (Program for Approximate Parameterization from Quan- tum Mechanical Data) which is developed to provide a tool for the determination of approximate bonded force parameters. In a following paper6’ the authors examine the reliability of the semi-empirical RHF (Restricted Hartree-Fock) wavefunction com- puted from MIND0/3,63 MND0,64 and AM 1 65 Hamiltonians to correctly represent the molecular characteristics in its perturbed geometries.Force field parameters derived semi-empirically are then compared to experimentally determined values used in the most popular force fields. An alternative to deriving parameters by optimizing agreement between experimen- 58 (a) N.L. Allinger Y.H. Yuh and J.H. Lii J. Am. Chem. SOC. 1989 111 8551; (b)J.H. Lii and N.L. Allinger J. Am. Chem. SOC. 1989,111,8566; (c)J. H. Lii and N. L. Allinger J. Am. Chem. SOC. 1989,111 8576. 59 (a)J. R.Maple U. Dinur and A. T. Hagler Proc. Nat. Acad. Sci. USA 1988,85,5350; (b)J. R. Maple T. S. Thacher U. Dinur and A.T. Hagler Chem. Design Aut. News.,1990 5(9) 5. 60 ‘DISCOVER 2.8’ Biosym Technologies Inc. 9685 Scranton Road San Diego CA 92121 USA. 61 C. Aleman E.I. Canela R. Franco and M. Orozco J. Comput.Chem. 1991 12 664. 62 C. Aleman and M. Orozco J. Cornput.-Aided Mol. Design 1992 6 331. 63 R.C. Bingham M. J. S. Dewar and D. H. Lo J. Am. Chem. SOC. 1975 97 1285. 64 M. J.S. Dewar and W. Thiel J. Am. Chem. SOC. 1971 99,4899. 65 M. J. S. Dewar E.G. Zoebisch E. F. Healy and J. J. P. Stewart J. Am. Chem. SOC. 1985 107 3902. Physical Methods and Techniques -Part (ii) Computer Graphics 29 tal and potential energy predictions is to compare experiment with free energy. The free energy perturbation (FEP) is given by AG = G -C; = -RTln<exp( -AH/RT)> (21 where AG is the free energy difference between states A and B AH is the difference between the Hamiltonians representative of these states R is the gas constant and T the temperature. Pearlman and Kollman66 use FEP in conjunction with a constraint method to generate torsion maps for comparison with potential energy torsion maps.The two maps for nucleosides are qualitatively similar but display significant quantitative differences. This indicates the significant role entropy can play in stabilizing various conformers. In the abstracts of the 204th American Chemical Society (ACS) National Meeting a number of works were presented concerning force field development. For instance recent developments of the Chem-X67 and MM368 force fields. Also given is the form scope parameterization and performance of the Merck Molecular force field.69 A systematic method for estimating MM2 parameters is given by Liu and P~rvis.’~ Cornell et al.” present a second-generation force field for proteins nucleic acids and small molecules.The development of a force field for modelling organometallics is presented by Gilbert et At earlier meetings the modelling of carbohydrates and polysaccharides was presented by brad^'^ and the derivation of QSAR parameters from quantum mechanical and force field calculations was presented by Hemken and Lehmann.74 For molecules in solution calculations are complicated due to the presence of solvent interactions which significantly affect conformational energies. An article by Teeter,75 for example discusses the theory and experiment behind water-protein interactions. Explicit inclusion of solvent molecules is practised but such an approach is computationally demanding. It is therefore desirable to have a potential function which accounts for solvent effects without the need to incorporate solvent molecules into the modelling system.Gilson and Honig76 have introduced a new pairwise energy term which accounts for charge-solvent interactions and can easily be incorporated into existing force fields. The POLARIS77 software which allows modelling of proteins in their solvent environment has been developed to allow free energy perturbation (FEP) calculations. BOSS7*is a program for performing Monte Carlo simulations for solutions containing a small number of solute molecules in either a solvent or in a dielectric continuum. 66 D. A. Pearlman and P. A. Kollman J. Am. Chem. Soc.. 1991 113 7167. ’’ K. Davies and M. Baird Ah. Papers.Am. Chem. Soc.. 1992 204. 41. ‘* J. P. Bowen P. C. Fox G. Y. Liang G. McGaughey. J. Y. Shim. and E. L. Stewart .4hs. Papers Am. Chem. Soc. 1992 204 39. 69 T.A. Halgren Ahs. Papers Am. Chem. SOL..,1992 204. 38. ’’ S.Y. Liu and G. D. Purvis Ahs. Papers Am. Chem. Soc. 1992. 204 33. 71 W. D. Cornell P. Cieplak I. R. Gould K. M. Merz J. W. Caldwell D. C. Spellmeyer. and P. A. Kollman Abs. Papers Am. Chem. Soc. 1992 204 40. 72 K. E. Gilbert J. J. Gajewski. and T. Kreek. Abs. Papers Am. Chem. Soc.. 1992. 204 379. 73 J. W. Brady Ah. Papers Am. Chem. Soc. 1992 203 4. 74 H.G. Hemken and P. A. Lehmann Ahs. Papers Am. Chem. Soc.. 1991. 202. 44. l5 M. M. Teeter Annu. Rev. Biophys. Biophys. Chem. 1991 20. 577. 76 M.K. Gilson and B. Honig J. Compur.-Aided Mol. Design 1991.5. 5. ” ‘POLARIS’ Molecular Simulations Inc. 200 Fifth Avenue Waltham MA 02154 USA. 78 ‘BOSS’ Tripos Associates Inc. 1699 South Hanley Road Suite 303 St Louis MO 63144 IJSA. 30 C.I. de Matteis D.E. Jackson and N. Raj An important feature in molecular modelling is the treatment of electrostatics. Interactions involving electrostatics are crucial in such cases as the understanding of conformations of solutes in highly dielectric solvents and also the recognition process of ligands by biological receptors. Most electrostatic calculations rely on the assumption that partial charges may be placed at atomic nuclei. These charges may be calculated rigorously by evaluating the Molecular Electrostatic Potential (MEP) from the wavefunction of a particular molecule in a given conformation.Fitting procedures are then used to assign partial charges so as to obtain the required MEP. Such an approach is clearly a function of molecular conformation and different MEP applications in drug design are discussed by Pepe et A technique for calculating partial charges given just initial nuclear coordinates is used in the CHARGE2 program. CHARGE2 is an empirical procedure for the rapid evaluation of partial charges with the original paper" concerned solely with haloalkanes. Since the first publication CHARGE has continued to develop and is now widely used in the modelling community. CHARGE2 is based on two fundamental chemical concepts the inductive effect in saturated molecules and Hiickel Molecular Orbital calculations for n-systems.The inductive effect operates uia the atomic electronegativity and polarizability and the Hiickel scheme works through appropriate Coulomb and resonance integrals. Parameterization of CHARGE2 is based on the observed molecular dipole moments. In a recent article" the authors consider the problems of including silicon together with methods of overcoming them. Also partial atomic charges for all the natural amino acids and the nucleic acids adenine cytosine guanine and thymine are given. In reference 82 the charge scheme of reference 80 is subjected to major re-parameterization in order to obtain a more general series of parameters describing the bonding interactions in saturated organic molecules. Mullays3 reports on a simple method for calculating atomic charges in charged molecular systems.The method is based on the orbital electronegativity (EN) concept and utilizes an EN equalization principal. Results show that the method compares well with high level theoretical calculations for a variety of charged molecules including alanine and protonated adenine. A recent development in modelling software is HYPERCHEM84 3D which offers four molecular mechanics force fields MM' AMBER BIO' and OPLS. MM' supplements the standard MM2 force field by providing additional parameters and BIO' is an implementation of CHARMm. This modelling system can be run on either 386/486 PC or UNIX platforms. Bays has given a broad overview of five molecular modelling programs for the Macintosh.85 The packages reviewed are Alchemy 11,86 Nemesis,87 PC-Model,88 79 G.Pepe D. Siri and J. P. Reboul J. Mol. Struct. (Theochem) 1992 88 175. 'O R.J. Abraham L. Griffiths and P.J. Loftus J. Cornput. Chern. 1982 3 407. " R. J. Abraham G. H. Grant I. S. Haworth and P. E. Smith J. Cornput.-Aided Mol. Design 1991 5 21. " R. J. Abraham and G. H. Grant J. Cornput.-Aided Mol. Design 1992 6 273. '3 J. Mullay J. Comput. Chern. 1991 12 369. 84 'HYPERCHEM' developed and licensed from Hypercube Inc. Supplied by C.E. Systems Unit 55 Suttons Park London Road Reading RG6 lAZ UK. *5 J. P. Bays J. Chern. Educ. 1992 69 209. '6 'Alchemy II' Tripos Associates Inc. 1699 South Hanley Road Suite 303 St Louis MO 63144 USA. 87 'NEMESIS' Oxford Molecular Ltd The Magdalen Centre Oxford Science Park Sandford-on-Thames Oxford OX4 4GA UK." 'PC-MODEL' Serena Software Box 3076 Bloomington IN 47 402-3076 USA. Physical Methods and Techniques -Part (ii) Computer Graphics 31 MacMimic,*’ and Chem3D P~US.’~ The review describes features such as how structures are constructed and then displayed and manipulated. The software also allows molecular mechanics computations with each package having its own force field. Alchemy I1 uses SYBYL; MacMimic uses MM2(87); PC-model and Chem-3D Plus force fields are based upon modified MM2; and Nemesis uses COSMIC-90. Molecular Simulations.-In energy minimization the task is to find the set of independent variables x = (x,,x,,x, . . .,x,,) for which the function V = V(x)has its minimum value.In the case of a molecular system comprising N atoms the 3N components of x are the atomic coordinates and Vis the potential energy. It should be pointed out that the problem of finding the global minimum of a general non-linear function with a realistic number of independent variables is extremely difficult. In molecular energy minimization the two algorithms most commonly used are steepest-descent and conjugate-gradient . Kini and Evans’ have reported on their work to develop a procedure which may routinely be used to build models of homologous proteins starting from the experimental structure of a closely related protein. Energy minimization strategies were performed on melittin and cardiotoxin by both constrained and unconstrained pathways. In addition the effects of the steepest-descent and conjugate-gradient algorithms for energy minimization were compared.Based on these results molecular modelling was applied to lysozyme mutants. In reference 92 a molecular mechanics energy minimizer ORAL is presented whose main features include ‘floating blocks’ and ‘isles’. The blocks are sets of atoms grouped together by the user and isles allow interactions between groups of atoms to be ‘switched-off ’. The program possibilities are presented by examples of molecular docking energy barrier estimation modelling of infinite structures and DNA bending simulations. Morley et al.’ have developed a hybrid conformational search algorithm Dynamic Monte Carlo (DMC) that combines a modified form of molecular dynamics with Metropolis Monte Carlo sampling.In DMC trial configurations are generated by short bursts of high-temperature dynamics in which the initial kinetic energy is focused into single bond rotations or ‘corner-flapping’ motions in ring systems. Constant temperature and simulated annealing protocols were then applied to conformational analysis of several model hydrocarbons. The work in reference 93 has since been extended to allow modelling of multi-fragment systems. This is achieved by incorporat- ing additional motions to focus energy into fragment translates and rotates. The complexes of macrocyclic hosts with suitable guest molecules are of great current intere~t,’~ both in their own right and also as model systems for studying molecular recognition and the binding interactions of larger systems such as enzyme-substrate complexes95 and macromolecule-water interaction^.^^ Molecular 89 ‘MacMimic’ Instar Software AB Research Park IDEON $22 370 Lund Sweden.9” Them-3D PLUS’ Cambridge Scientific Computing Inc. 875 Massachusetts Avenue Suite 41 Cambridge MA 02139 USA. 91 R. M. Kini and H. J. Evans. J. Biomol. Struct. Dyn. 1991 9 475. 92 K. Zimmermann J. Comput. Chem. 1991 12 310. 93 S. D. Morley D. E. Jackson M. R. Saunders. and J. G. Vinter J. Comput. Chem. 1992 13 693. 94 (a)J. M. Lehn Angew. Chem..Int. Ed. Engl. 1988,27 89; (6) D. J. Cram. Angew. Chem..Int. Ed. Engl. ’’ (a) 1988 27 1009. J. Blaney P.Weiner. A. Dearing P. Kollman E. Jorgensen,S. Oatley,J. Burridge,and C. Blake J. Am. Chem.Soc. 1982,104,6424; (6)G.Wipff A. Dearing P. Weiner J. Blaney and P. Kollman J. Am. Chem. Soc. 1983 105. 997. 32 C.1.de Matteis D. E. Jackson and N. Raj modelling has been applied to host-guest or receptor-ligand complexation to study the influence of steric-fit and electrostatic interactions. A recent article by Harris96 gives an interesting account of the nature of 'Inclusion Complexes' in which two types of confinement are described. In one type the host is the molecule with a form of cavity able to enclose a suitable guest. In the other type of complex the guest is embedded within well-defined cavities of a crystalline host. Hart and Read9' present the Multiple-Start Monte Carlo Docking Method to search for possible binding modes of molecular fragments at a specific site of a potential drug target.An MC run is first performed in which the energy in the Metropolis algorithm is replaced by a score function that measures the average distance of the probe to the target surface. This has the effect of making buried probes move toward the target surface and allows enhanced sampling of deep pockets. In the second stage an energy-driven MC is used to recover all favourable states of the bound complex. Protein-protein docking simulations have been reported by Cherfils et Anti-body-lysozyme and protease-inhibitor complexes are reconstituted by docking lysozyme as a rigid body onto the combining site of the antibodies and the inhibitor onto the active site of the proteases. The proteins are modelled with one sphere per residue and subjected to simulated annealing using a crude energy function.Five out of six of the resulting complexes retained the X-ray crystallography structures. A further docking algorithm based on graph theoretical techniques is presented by Kasinos et Results are given for the molecular recognition problem where given information concerning particular atoms involved in the binding for one molecule the algorithm correctly identifies the corresponding atoms of the approaching molecule. Gussio et a/.'00point out the importance of dielectric effects in their study of the docking of 4,5-a-epoxymorphinans into an Asp-Lys-His-Phe pseudoreceptor. Kostense et a!. report on modelling of B-cyclodextrins hosts with a variety of guest molecules by rigid body docking.Inclusion of a guest inside the cyclodextrin cavity causes the cavity to elliptically distort and the amount of distortion is related to the van der Waals volume of the guest. This enabled the authors to develop a procedure to construct fi-cyclodextrin molecules that are able to encompass guest molecules with a given van der Waals volume. Morley et a!.'02 have performed a number of complexation studies using their multi-fragment DMC algorithm. The simulations were performed with no constraints applied to host or guest molecules. Successful docking with good agreement with crystallographic data was obtained for p-tert-butylcalix[4]arene with toluene and dibenzo-34-crown-10 with [Methyl viologen12 + . In addition modelling results are presented for inclusion complexes studied experimentally by Cram et ~1."~of crown ether hosts with chiral amino acid-derivative guests.96 K. D. M. Harris Chem. Br. 1993 29 132. 97 T. M. Hart and R. J. Read Proteins Struct. Funct. Gene. 1992 13 206. 98 J. Cherfils S. Duquerroy and J. Janin Proteins Struct. Funct. Gene. 1991 11 271. 99 N. Kasinos G. A. Lilley N. Subbarao and I. Haneef Prot. Eng. 1992 5 69. *Oo R. Gussio S. Pou J.H. Chen and G. W. Smythers J. Cornput.-Aided Mol. Design 1992 6 149. lo' A. S. Kostense S.P. van Helden and L. H. M. Janssen J. Cornput.-Aided Mol. Design 1991 5 525. lo' S.D. Morley. N. Raj D. E. Jackson and P. M. Williams in 'Computer Aided Innovation of New Materials II' ed. M. Doyama J. Kihara M. Tanaka and R. Yamamoto Elsevier Amsterdam 1993 p.835. lo3 (a)D. J. Cram R.C. Helgeson S.C. Peacock L. J. Kaplan L.A. Domeier P. Moreau K. Koga J. M. Mayer Y. Chao M.G. Siege] D. H. Hoffman and G. D. Y. Sogah J. Org. Chem. 1978,43 1930; (b)C. B. Knobler F. C. A. Gaeta and D. J. Cram J. Chem. Soc. Chem. Cornmun. 1988 330. Physical Methods and Techniques -Part (ii) Computer Graphics 3 Chemical Databases The number and size of chemical information databases continues to increase together with the development of more sophisticated search and retrieval software. ‘Beilstein’s Current Facts in Chemistry’ is a database of chemical literature which can be queried using a number of data fields and is available on CD-ROM.lo4 ‘Comprehensive Heterocyclic Chemistry’ is a database of reactions involving heterocyclic compounds and can be searched by 2D structure or substructure and by a variety of textual data fields.lo5 ‘Beilstein Online’ now contains over 6.2 million preparations and reac- tions,lo6 whilst the Chemical Abstracts Registry database now contains 4.5 million substances.lo’ Chemical Abstracts is developing a new information service where concept searching and browsing facilities will be available.Text and pictures will be integrated and Hypertext navigation will allow easy access to related textual information at the touch of a button.”* The increasing availability and use of three dimensional structural data from experimental and theoretical techniques has fuelled the development of new databases and the expansion of existing ones allowing the storage and retrieval of three dimensional information.The Cambridge Structural Database now contains atomic coordinates for more than 100000 organic and organometallic compounds and 10000 new additions are predicted annually. The version V database allows searches by 3D and 2D structures and substructures together with intramolecular and intermolecular sear~hes.’~’ The Brookhaven Protein Data Bank contains three dimensional atomic coordinates for proteins nucleic acids and polysaccharides and the number of complete structures in January 1993 was 1055. The database cannot be searched at present although developments are underway to develop more intelligent retrieval software.’ lo ‘Iditis’ is a database that stores structural data calculated from Brook- haven Protein Data Bank structures and allows searching through these data for structural features of interest.In total 370 different fields of information are calculated for each protein including descriptions of motifs secondary structure amino acid and atomic structure positions of hydrogen bonds and positions of disulfide bridges.’ CAST-3D is an online database developed by Chemical Abstracts containing the three dimensional coordinates for a growing number of substances within the ‘Chemical Abstracts Registry File’. The atomic coordinates have been obtained using CON-CORD. This database can be searched three dimensionally and using Registry numbers for the compounds located allows direct querying of other Chemical Abstracts databases.’ l2 SYBYL/3DB UNITY is a commercially available three ‘Beilstein’s Current Facts in Chemistry’.Springer-Verlag Electronic Media Department 175 Fifth Avenue New York NY 10010 USA. In5 ‘Comprehensive Heterocyclic Chemistry’ Molecular Design Ltd. 10Armstrong Mall. Southwood Summit Centre Farnborough Hampshire GU14 ONR UK. Io6 ‘Beilstein Online’ Springer-Verlag Electronic Media Department 175 Fifth Avenue New York NY 10010. USA. lo’ ‘CAS Registry System’ Chemical Abstracts Services. Io8 M. Withers Chem. Br. 1993 29 274. F. H. Allen and 0.Kennard Chem. Des. Auto. News. 1993 8 1. ‘Io ‘Protein Data Bank’ Chemistry Department Building 555 Brookhaven National Laboratory Upton NY 11 973 USA. ’ ’ ‘Iditis’ Oxford Molecular Ltd. The Magdalen Centre Oxford Science Park Sandford-on-Thames Oxford OX4 4GA UK.‘I2 ‘CAST-3D’ Chemical Abstracts Services. C.I. de Matteis D.E. Jackson and N. Raj dimensional data storage and retrieval system which can import data from a number of commercial and public domain databases together with storing in-house data.’ l3 Pharmastructures is a PC based 3D structure database at present containing 4000 structures.’ l4 Further MDL and ICI are developing software which will incorporate conformationally flexible searching into the 3D search software MACCS-3D and ISIS/3D. This is achieved by allowing rotation about single bonds to create the conformation necessary to match a query.’ ‘I3 ‘SYBYL-3DB UNITY’ Tripos Associates Inc. 1699 South Hanley Road Suite 303 St Louis MO 63 144 USA.‘Pharmastructures’ 18-20 Hill Rise Richmond Surrey TW 10 6UA UK. ‘I5 B.F. Graham Chern. Des. Aut. Nrris. 1993 8 30.