Abstract Listing

Abstract
The docking program, GOLD, has been further developed in collaboration with Astex Technology to allow users to dock ligands while accounting for the presence or abscence of mediating water molecules in binding.

Preliminary validation shows a good rate of prediction of correct water-mediated binding modes. In this talk, the methodology used will be presented along with results of further validation using a larger test set.

CCDC has also developed a tool to allow users to post-process docking results to calculate customizable protein-ligand interface descriptors. The tool utilises an XML driven command language for the intersection of primary properties to form compound properties. This work will be presented.

Abstract
Economic pressures have intensified efforts to decrease the time needed to bring a new drug to market and to reduce the cost, and computational methods are increasingly relied upon to expedite lead discovery. We have recently made fundamental advances in sampling and scoring algorithms for virtual screening that have led to greater accuracy in identifying active ligands. In particular, improved docking accuracy has enabled us to develop an advanced scoring function ("extra precision" mode, XP) for Glide (J. Med. Chem. 2004, 47, 1739-1749; 1750-1759) that effectively rejects "false positives" by significantly penalizing physically inappropriate interactions and that more efficiently identifies compounds likely to bind strongly by rewarding key binding motifs. The increased enrichment allows a smaller number of docked ligands to be submitted to further detailed analysis, including assessment of induced fit, ligand strain energy and visualization. Using our hierarchical protocol we docked a database of 500,000 available compounds, prefiltered for drug-like ADME properties with QikProp, to a p38 MAP kinase receptor conformation believed to favor selectivity. Of 1000 top-scoring compounds, post-processing produced a shortlist of 70 for purchase and screening as kinase inhibitors. Of 28 compounds screened to date, 8 compounds representing novel chemotypes exhibited better than 50% inhibition and may be suitable as lead candidates. Four compounds have been analyzed further and gave IC₅₀ values ranging between 1.8 and 9.8 µM; IC₅₀ determinations are pending for the remaining 4 hits, but we believe that one or two of these may be submicromolar. This study demonstrates that our computational algorithms and our virtual screening methodology are mature enough to present a viable alternative to High Throughput Screening at greatly reduced cost where sufficient target structural information is available.

Abstract
Docking algorithms seek to extremize the value of a scoring function by finding the optimal conformation and alignment (pose) of a ligand relative to a protein binding site. The Surflex docking method employs an empirically constructed scoring function and a search methodology that makes use of molecular similarity based pose generation. Comparative results will be presented on both docking accuracy (geometric agreement with crystallographic experiment) and screening enrichment (ability to distinguish true positives from false positives). Particular emphasis will be on screening efficiency and methods to improve scoring functions by explicit training using putative false positives. Performance of Surflex v1.21 is competitive with many popular docking methods for docking accuracy and appears to be significantly better for screening enrichment.

Abstract
We describe a new method for modeling protein and ligand main-chain flexibility in docking. The goal is to sample the full conformational space, including conformations not yet observed by crystallography, MD, or NMR. Flexibility analysis is performed using the graph-theoretic algorithm FIRST, which identifies coupled networks of covalent and non-covalent bonds within the protein. ROCK then explores available conformations by only sampling dihedral angles that preserve the coupled bond network in the protein. A representative set of protein conformations can then be used as targets for docking with SLIDE, which models protein and ligand side-chain flexibility. This combined approach for incorporating main-chain flexibility in docking is illustrated for cyclophilin A-cyclosporin and estrogen receptor-zearalenol complexes. Very recent results show that the maintenance of correlated motions between hydrogen-bonded and hydrophobic side chains is also a key aspect of ligand recognition across diverse protein-ligand complexes.

Abstract
MVP is a molecular mechanics program with facilities for docking, conformational search, library enumeration and homology modelling (Lambert, "Docking Conformationally Flexible Molecules into Protein Binding Sites," in Practical Application of Computer Aided Drug Design, Charifson, ed, (1997)). MVP was originally developed for protein structure prediction (Lambert and Scheraga, JCC 10 770-797, 798-816, 817-831, (1989)), and implements a generalized version of Harold Scheraga's build-up procedure. MVP can dock flexible organic molecules into a protein by running this build-up, or "grow" calculation within the binding site. This build-up process requires that the compound have an "anchor group" with approximately known position and orientation in the binding site. The anchor group is subjected to limited rotations and translations, but the procedure cannot usually predict large shifts in anchor position or orientation. These MVP docking calculations have been used in the structure based design of numerous compounds at Glaxo SmithKline, including eight clinical candidates, two of which are now in Phase II clinical trials. We will illustrate how the MVP build-up process works, and describe recent improvements, including methodology that makes it possible to dock compounds without any anchor group.

Abstract
Scoring is the limiting factor in well designed scoring programs. Several several scoring functions are evaluated using the FRED docking program. Functions are evaluated again two criterion. First, their ability to pick out a correctly docked structure from many alternate poses. Second, their ability to select ligands with high binding affinity from a set of decoys. Functions examined include Zap Bind (a PBSA based scoring function), Chemgauss (a smooth chemically aware gaussian based scoring function) and the well know Chemscore and PLP scoring functions. Several other functions will likely be added by the time of this presentations.

Abstract
This presentation discusses conformational Genetic Algorithms. The conventional Gray Coding of torsional angles of molecules in pharmacophore elucidation tool GASP was substituted with a Biased Gray Coding (BGC). BGC explores torsional angles of molecules close to certain predetermined values where one is at better odds of finding a good solution. Results of GASP with BGC are compared with conventional GASP Gray Coding results.

Abstract
Monomers used as dental restorative have the potential to leach out of the resulting polymer matrices due to incomplete polymerization, and may thus enter the human blood stream. Such materials must be evaluated very carefully for various toxicity effects. This study will focus on the prediction of mutagenicity (as defined by mutation of the standard Ames TA100 strain of bacteria) of several potential dental materials through quantum mechanically-based quantitative structure activity relationships (QM-QSARs). The SAM1 semiempirical method is used in this study allowing extension and refinement of our previous work to include new silicon-containing dental monomers. Also, a new mathematical interpretation of the TA100 data was used to generate QSARs based on different property values than those employed previously. Comparison of the new and previously developed QSAR models will be reported.

Abstract
We wish to report on the nature and strength of fluorine-mediated intermolecular forces. Binding energies between representative fluorinated molecules and cations / hydrogen bond donors have been calculated through high level ab initio calculations extrapolated at the complete basis set limit. NBO analysis is then used to provide an explanation for the origin of these interactions. The role of the degree of substitution on the carbon atom bearing a fluorine functionality has also been analyzed. We will provide evidence of the change in the ability of carbon-bound fluorine to engage in intermolecular interactions based upon the nature of the groups attached to the fluorine-bound carbon atom. Quantitative differences in binding abilities of mono-, di-, and trifluoromethyl functionalities will also be discussed along with corresponding applications in drug design and discovery.

Abstract
To produce low cost fullerenes by high-energy lasers, structure, stability, and nucleation of fullerene clusters are considered and related to the optimal control formulation at the molecular scale. Based upon indications that the behavior of the plasma/He interaction volume for laser ablation experiments essentially follows that for cathodic arc discharge experiments, the possible zones of formation of the large carbon molecules and the feasibility of a more efficient fullerene synthesis by lasers are discussed. We propose a computational algorithm linking fullerene yield and production rate to laser characteristics. The optimized plasma zones providing a C₆₀ yield of 70% and a production rate of 5.14 g/min (P_He = 240 Torr; power density = 7×10⁹ W cm^-2, T = 2700 K) from a graphite target are identified and displayed as time response. Cathodic arc systems are suggested to be valuable tools in the determination of local conditions for fullerene formation.

Abstract
The ability of C-H bonds to participate in hydrogen bonding has been an issue of long-standing interest. In the literature, the proposed nature of the interaction between substituted methanes RCH3 and halide ions ranges from purely electrostatic, ion-dipole attraction to regular or improper, blue-shifting hydrogen bonding. The situation is especially unclear in the case of halide-acetonitrile complexes and clusters, since several reported theoretical and experimental studies disagree on the prevalent role of either ion-dipole interactions or hydrogen bonding. In this contribution, we will present a detailed systematic computational study of the structures, binding energies and potential energy surfaces for the series of small halide-acetonitrile clusters X^–(CH₃CN). The applicability of various quantum-chemistry methods to the systems of interest will be discussed and the nature of the bonding interactions in the X^–(CH₃CN) complexes will be analyzed using the Atoms-In-Molecues and Natural Bond Orbitals approaches and various energy decomposition schemes. In light of this computational study, possible explanations for recent cluster and bulk solution experimental results and existing theoretical models will be critically discussed.

Abstract
Computational chemistry has expanded from computation of hydrogen molecule to computation of living cell. Applying theories from ab initio quantum mechanics to various simplified models, the virtual worlds explored by computations provide different replicas of real world phenomena. What kind of mapping relationships should we expected from our studies? How do we interact with experimental information? The computational biology is in the world with complex organization, for which a unified theory is yet to be proposed. A computational biological model, even with clear physical or chemical meanings, may not be necessarily reduced to physics or chemistry. One common theme from computational quantum chemistry to computational biology is that the virtual worlds can affect our perception of real world. To make the perception to be truth, we have to increase mutual interaction of computation and experiment.

Abstract
We present our development of an integrated extendable Web-based Grid-computing environment for computational science and engineering called Computational Science and Engineering Online (CSEO). CSEO allows computational scientists to perform research using state-of-the-art tools, to query data from personal or public databases, to document results in an electronic notebook, to discuss results with colleagues, and to access grid-computing resources from a web browser regardless of geophysical location or time zone. Currently, CSEO provides an integrated environment for multi-scale modeling of complex reacting systems and biological systems. A unique feature of CSEO is in its framework that allows data to flow from one application to another in a transparent manner. Advantages, disadvantages, and future prospects of CSEO are then discussed. CSEO can be accessed at http://cseo.net.

Abstract
Gas-phase reactions with ozone are significant removal process for volatile unsaturated hydrocarbons in troposphere, which also produce atmospherically important species: HO, HO₂, organic and inorganic peroxides. The primary product of the ozonolysis of alkenes is the Criegee intermediate (CI), the most important reaction of which is interaction with water. In present work ab initio and density functional methods were applied to evaluate importance of formation of CI complexes with water clusters for atmospheric reactions. Various structures CI…(H₂O)_n with n=1..4 were calculated and minimal configurations are found. In addition, the reactions of these complexes were investigated; energy barriers and rate constants were estimated. The rate constant of overall process was calculated, and its dependence on temperature and relative humidity was determined. Reaction rates for all pathways were estimated based on found rate constants and typical concentrations of the reactants in atmosphere. The further reactions of products and its impact on the chemistry of atmosphere are discussed.

Abstract
Glycosidases and lectins both bind sugars, but only the glycosidases are catalytic. The glycosidases occur among 90 evolved protein families. Family 18 is one of the two familes of chitinases (EC 3,2.1.14). Interestingly, lectins are also in this evolutionary group of Family 18 glycosidase proteins. Proteins belonging to the enzymatically inactive class ("chitolectins") have a highly similar binding site to the catalytic Family 18 enzymes. One major exception is a glutamic acid which acts as the essential acid/base residue for chitin cleavage is replaced with leucine or glutamine. We present our comparison of the recently obtained structures of two Family 18 chitolectins, MGP40 (Mohanty, Singh et al., 2003) and GP39 (Fusetti, Pijning et al., 2003; Houston, Anneliese et al., 2003).

Abstract
Chiral diamine-silver I complexes have shown chiral recognition abilities toward chiral alkenes [1]. We are investigating chiral dinitrogen ligands (diamine, 1,10-phenanthroline)-silver I complexes and their binding to alkenes and alkynes using both quantum and molecular mechanics. Ab initio calculations were used to develop additional parameters for the MM3 force field for the accurate description of the geometry and relative energies of the transition metal containing complexes (Q2MM method [2]). The application of this force field to the prediction of the best ligand for a high enantiomeric excess in the resolution of racemic alkenes or alkynes will be discussed.

[1] Organometallics 2004, 23, 15-17 [2] J. Mol. Struc. (Theochem) 2000, 506, 9-16

COMP 17 [774178]:  Nanoscale manipulation of hydrogen storage in NaAlH4: Exploring catalytic surfaces using density functional theory
Santanu Chaudhuri, Nanocatalysis Group, Department of Chemistry & Center for Functional Nanomaterials, Brookhaven National Laboratory, Building 555, Upton, NY 11973, chaudhuri@bnl.gov, Ping Liu, Dept. Chemistry, Brookhaven National Laboratory, and James T. Muckerman, Chemistry Department, Brookhaven National Laboratory

Abstract
NaAlH4 doped with ~2% titanium is a promising hydrogen storage material. Density Functional Theory using the RPBE functional can predict the role of Ti during the multi-step hydrogen absorption-desorption cycle. Two of the most probable mechanisms of Ti assisted hydrogen storage e.g. the replacement of Na by Ti on the surfaces and formation of a Ti-Al alloy, have been probed in this work. The energetics of the hydrogen absorption process indicates that the intermediate perovskite phase, Na3AlH6, is less reactive compared to the end product of the hydrogen desorption cycle, NaH and Al. The NaH surface doped with Ti (figure 1) has been found to promote exothermic dissociative absorption of molecular hydrogen. This explains why nanometric NaH doped with Ti is reported to be a good hydrogenation catalyst. The use of DFT in unraveling the myriad correlations between electronic structure, oxidation state, defects and hydrogen storage efficiency will be discussed.

COMP 18 [773389]: DFT study of the interactions of antiwear additives with iron and iron oxide
Hongmei Wen, Susanne M. Opalka, and Clark V. Cooper, Physical Sciences Department, United Technologies Research Center, 411 Silver Lane, East Hartford, CT 06108, wenh@utrc.utc.com

Abstract
Consensus methods[1-4] combine various scoring functions for in silico screening of large chemical databases against protein targets. They have been demonstrated to provide improved accuracy in the docking procedure, both the prediction of binding conformations and relative binding energies. We address here the issue of determining which scoring functions should be combined to obtain optimal results from the docking procedure. We illustrate this with examples where including too many scoring functions leads to a reduced hit-rate than when a select set of functions are used, and present a method of determining the best combination of functions to apply without the use of previously known ligands. The scoring functions used in the current analysis include DOCK, AutoDock, PMF, ChemScore, Score, SmoG. and X-cscore functions.

1. Bissantz C, Folkers G, Rognan D. Protein-based virtual screening of chemical databases. 1. Evaluation of different docking/scoring combinations. J Med Chem 2000 Dec 14;43(25):4759-67 2. Wang R, Wang S. How does consensus scoring work for virtual library screening? An idealized computer experiment. J Chem Inf Comput Sci 2001 Sep-Oct;41(5):1422-6 3. Wang R, Lai L, Wang S. Further development and validation of empirical scoring functions for structure-based binding affinity prediction. J Comput Aided Mol Des 2002 Jan;16(1):11-26 4. Clark RD, Strizhev A, Leonard JM, Blake JF, Matthew JB. Consensus scoring for ligand/protein interactions. J Mol Graph Model 2002 Jan;20(4):281-95

Abstract
Scoring functions for virtual screening are usually optimized to work for diverse sets of compounds. The question we wanted to answer is whether it is possible to improve the performance of the scoring function once a few active ligands have been identified. In receptor docking, the scoring function has two utilities: ligand placement in the pocket and ranking docked solutions. In our work we separated the two tasks: the former was left to the docking program, while we developed a novel function for the latter role. This function is based on the buriedness of the ligand in the receptor pocket, possible hydrogen bonding interactions and calculated binding energy. Receptor buriedness is a measure of how well molecules occupy the binding pocket in comparison to known high-affinity ligands. The possibility of hydrogen bond formation is checked for selected residues that are recognized as being important in the binding of known ligands. The approximate binding energy is calculated from the thermodynamic cycle. The information necessary for the scoring function can ideally be gleaned from the 3D structure of the receptor-ligand complex or the 3D structure of the receptor and known active ligands that bind to the given site. We show that the new scoring functions provide up to 12 times improvement in enrichment compared to the popular commercial docking program GOLD.

Abstract
Knowledge-based potentials have been found useful in a variety of biophysical studies of macromolecules. Recently, it has also been shown in self-consistent studies that it is possible to extract quantities consistent with pair potentials from model structural databases. In this study, we attempt to extend the results obtained from these self-consistent studies toward the extraction of realistic pair potentials from the PDB. The method utilizes a clustering approach to define particle types within the PDB consistent with the optimal effective pairwise potential. The method has been integrated into the SMoG drug design package, resulting in an improved approach for the rapid and accurate estimation of binding affinities from structural information. Using this approach, it is possible to generate simple knowledge-based potentials that correlate strongly (R=0.61) with experimental binding affinities in a database of 118 diverse complexes. Further, predictions performed on a random 1/3 of the database consistently show an average unsigned error of 2.5 log Ki units. It is also possible to generate specialized knowledge-based potentials, targeted to specific protein target families. This approach is capable of generating potentials that correlate very strongly with experimental binding affinities within these families (R=0.8 0.9). Predictions on 1/3 of these family databases yield an average unsigned errors ranging from 1.2 to 1.3 log Ki units. Altogether, we describe a physically motivated approach to optimizing knowledge-based potentials for binding energy prediction that can be integrated into a variety of stages within a lead-discovery protocol.

Abstract
Virtual high throughput screening, and lead optimization are very different problems. General docking tools can aid in both cases, but lead compounds or series are often accompanied by structure-activity information. Using structure-activity relationships (SAR) to direct ligand-protein docking can lead to higher quality binding-mode hypotheses. We will describe pose generation and evaluation algorithms that utilize SAR to guide their behavior. These methods will be evaluated across a variety of ligand-protein systems in the context of a lead-optimization docking tool with an MMFF-PB/SA binding potential.

Abstract
The cleavage of β-amyloid precursor protein (APP) by β-Secretase (BACE) is a crucial step in the production of the β-amyloid peptide that has been implicated as a probable cause of Alzheimer’s disease (AD). This has made BACE an attractive therapeutic target for treatment of AD. There are two aspartic acid residues (Asp 32 and Asp 228) in the catalytic region of BACE that can adopt multiple protonation states and tautomers. The protonation state and precise location of the protons for these two residues, particularly in the presence of an inhibitor, have a direct bearing on efforts to model this system properly. In the present study, we have carried out full quantum mechanical calculations using a linear scaling quantum mechanical method to identify the preferred protonation states and proton locations for Asp 32 and Asp 228 in the presence of inhibitors. Additionally, a binding affinity model based on the LIE approach has been developed that is capable of rank ordering inhibitors of BACE.

Abstract
β-Secretase (also known as β-APP Cleaving Enzyme or BACE-1) is one of two proteases responsible for processing the membrane-bound Amyloid Precursor Protein (APP) to the 40/42 residue β-amyloid peptide (Aβ), the primary constituent of the amyloid plaques observed in the brains of Alzheimer’s patients. Since BACE-1 cleavage of APP appears to be the rate-limiting step in the production of Aβ and BACE-1 knockout mice show complete absence of Aβ with no reported side effects, BACE-1 appears to be an attractive therapeutic target in the treatment of Alzheimer’s disease. BACE-1 has been characterized as the first known example of a pepsin-like aspartyl protease that is membrane-tethered. However, a crystal structure of the soluble domain reveals a high degree of similarity to the tertiary structures of other mammalian and fungal aspartyl proteases, e.g. renin, cathepsin D, and endothiapepsin. Given the availability of 3D coordinates for BACE-1, it appeared likely that an appropriate docking/scoring protocol could be identified which would aid in the design of BACE-1 inhibitors. Several scoring functions were evaluated based on the structures and observed activities for a small series of hydroxyethylamine inhibitors. To test the predictivity of the scoring, a virtual ‘reagent scan’ was performed to evaluate the predicted binding energy of approximately 700 amine reagents in the S1’ site. Several high-scoring amine reagents were selected for incorporation and led to potent BACE-1 inhibitors. This study demonstrates the utility of a virtual approach to selecting reagents. In addition, it supports previous qualitative conclusions about the character of the S1’ site in BACE-1 relative to other aspartyl proteases.

Abstract
Docking methods often generate small molecule poses that are close to crystallographically determined binding positions. Even though much effort has gone into improving scoring functions, existing scoring functions typically do not rank order candidate poses correctly. We present a complementary approach in which experimental results are used to generate models for activity that are then used to select among docking poses. We investigate whether ligand positions that lie at this intersection of two manifolds (the docking conformational space and experimentally-informed pharmacophore and shape spaces) improve compound selection by available scoring functions or by their application alone. We will present examples of this method applied to both public soluble ligand datasets and internal datasets obtained from several protein targets using Tethering^SM.

Abstract
Linear R12 methods of Kutzelnigg and coworkers make accuracies of better than 1 kcal/mol computationally feasible and routine. Initial applications of these methods were expensive due to the use of large MO bases to approximate or eliminate some many-electron matrix elements. The ABS MP2-R12 method, first studied by Klopper and Samson, uses a separate basis set for the approximate resolution of the identity. Since standard approximation of many-electron matrix elements assumes exactness of some MOs in Hartree-Fock sense, the question still remains: how complete the orbital basis set has to be? Here we investigate basis set convergence of some relative energies with respect to the orbital basis using the new scalable implementation of the MP2-R12 method in the publicly-available MPQC package. Preliminary results indicate that the MP2-R12 method in present form cannot be used safely when small basis sets (such as aug-cc-pVDZ and aug-cc-pVTZ) are utilized in the orbital expansion.

Abstract
We have implemented a Gaussian basis-set two-component self-consistent field method based on the fourth order Douglas-Kroll-Hess approximation. This variational two-component approach takes the spin-orbit interaction into account by employing a generalized Kohn-Sham scheme and allows one to deal with hybrid density functionals and open-shell systems. We present benchmark results in diatomics for equilibrium bond lengths, harmonic vibrational frequencies, and dissociation energies using local spin-density, generalized gradient, and hybrid density functionals. We also present results for the bond dissociation energies of uranium fluorides.

Abstract
Density functional theory (DFT) is often criticized for lacking a mechanical prescription for systematic convergence to the right answer. Nonetheless, performance of DFT can be gradually improved by imposing known analytic properties of the exact exchange-correlation functional on semi- and nonempirical approximations. We show how this approach is applied to constructing novel density functionals and illustrate its success with examples from atomic, molecular, and solid-state chemistry.

Abstract
The correlation consistent basis sets have played a pivotal role in enabling a hierarchy of high accuracy ab initio approaches to be well established. Though the sets have been widely used for main group species, their utility in transition metal studies has not yet been established, due to the very recent development of the sets. We have used the new sets to examine the impact of method, pseudopotential, and basis set choice upon the bonding description, and energetic and spectroscopic properties of a range of ground and excited-state transition metal species.

Abstract
We present a general scheme for evaluating one- and two-electron integrals using Fourier transformed expressions of Nuclear Attraction Integrals (NAIs) and Electron Repulsion Integrals (ERIs). First-order and second order derivatives of NAIs have been used to evaluate integrals associated with first-principle computation of electric fields and electric field gradients of molecular systems. In addition, first order and second order derivatives of ERIs are used in the evaluation of integrals associated with the calculation of zero-field splitting (ZFS) parameters. Higher order derivatives of NAIs and ERIs are not commonly used, however, our expressions could be used to arbitrary order should these have some practical application. Tables of formulas have been created to speed-up electronic structure calculations for the case of first and second order derivatives of NAIs and ERIs. Our work has direct application to the ab-initio calculation and interpretation of spectroscopic parameters generated by Mössbauer, EPR and magnetic susceptibility experiments.

Research supported by NSF grant CHE-0349189 (JHR).

Abstract
Abstract text not available.

Abstract
We study the electronic structure and bulk properties of UO₂ and PuO₂. Hybrid density functionals, Gaussian-type orbitals and relativistic effective-core potentials are used in a periodic boundary condition code. Such calculations for f-element solids were first reported in our paper on UO₂ [K. N. Kudin, G. E. Scuseria and R. L. Martin, Phys. Rev. Lett. 26, 266402 (2002)], where the established Perdew-Burke-Ernzerhof (PBE0) hybrid density functional was used. In the present work we perform similar calculations on PuO₂. Traditional density functionals have also been employed in both studies and they are found to compare worse with experiment, most likely due to the inadequate description of the localized f orbitals in actinide ions. The magnetic behavior was explored and PBE0 predicts the antiferromagnetic state to be the lowest in energy at T = 0 K, but nearly degenerate with the ferromagnetic state. The calculated lattice constant (5.39 Å) agrees very well with the experimental value of 5.40 Å, and PuO₂ is correctly predicted to be a small-band gap insulator. Inclusion of an interstitial oxygen atom in the PuO₂ lattice improves the agreement with the experimental density of states. We also compare the PBE0 results for both UO₂ and PuO₂ with data obtained with a newly developed hybrid density functional [J. Heyd, G. E. Scuseria and M. Ernzerhof, J. Chem. Phys. 118, 8207 (2003)].

Abstract
C-H bond cleavage is an important aspect in the combustion of organic fuels, such as coal. The homolytic C-H bond dissociation enthalpies, ΔH(298 K), of a series of substituted benzenes (substituents = -F, -Cl, -CN, -NO₂, -OH, -CH₃, -CF₃, -OCH₃, -SCH₃, -SH, -COH, -NH₂) were determined using hybrid density functional theory methods. Results showed a small substituent effect on the BDE of the C-H adjacent to the substituent and very little effect on the other C-H bonds in the ring. These values were also compared to those calculated in the corresponding substituted vinyl systems. Spin densities and the changes in electronic distribution were also evaluated using Bader’s theory of atoms-in-molecules. Furthermore, the reactivity under both atmospheric and combustion chemistry conditions of the substituted ethylenes with OH radical and H atom will be presented. Reaction barriers and energies were determined using hybrid density functional theory and composite ab initio methods.

Abstract
Remarkably accurate scaling relations are shown to exist between the correlation energy contributions from various excitation levels of the configuration interaction (CI) wavefunction, considered as functions of the size of the correlationg orbital space. These relationships are used to develop a new method for extrapolating a sequence of smaller CI calculations to the full CI energy. As a result the method also offers a systematic way for constructing compact and accurate CI wavefunctions. The method called correlation extrapolation by intrinsic scaling (CEIS) has been applied to neon atom and H2O, C2, N2, O2 and F2 molecules yielding the correlation energies of the benchmark quality.

Abstract
The gas-phase basicities (GBs) and the proton affinities (PAs) of all six isomers of dihydroxybenzoic acid (x,y-DHB), which are well known matrices used in matrix-assisted laser desorption/ionization mass spectroscopy, have been calculated using density functional theory (DFT) and Moller-Plesset perturbation theory. Respectively, the GBs vary from 803.9 kJ/mol for the least basic species (3,4-DHB) to 830.0 kJ/mol for the most basic one (2,4-DHB). The reported GBs and PAs are in good agreement with previous experimental measurements. The results indicate that protonation, in all six isomers, takes place on the carbonyl oxygens.

Abstract
We and others have recently demonstrated that nucleophilic addition to alpha,beta-unsaturated carbonyl compounds can be directed toward the alpha carbon rather than toward the beta carbon (Michael or conjugate addition) or the carbonyl carbon. This can be accomplished by attaching a pi-deficient ring or a phenyl ring multiply substituted with strong electron withdrawing groups at carbon beta. Such addition is also implicated in the nucleophilic addition to alkynoates catalyzed by trialkylphosphorus compounds. Such reactions have potential use for the synthesis of non-natural amino acids. We present a theoretical and experimental study of such reactions, identifying the barriers to reaction and the reaction routes with density functional and ab initio methods. We identify common features of these types of reaction.

Abstract
Combining theoretical and experimental approaches to determining rotational barriers for methyl and t-butyl groups in molecular solids leads to a more comprehensive understanding of the intermolecular interactions among molecules packed in a crystal environment. We present here results of ab initio calculations at the HF/6-311+G(d,p)//HF/6-31G(d) for the reorientation barrier for methyl and t-butyl group in four aromatic compounds, for which the crystal structures have been determined and the dynamics of methyl and t-butyl group motions have been measured by low frequency NMR relaxometry: 1,4-di-t-butylbenzene, 2,6-di-t-butylnaphthalene, 2,6-di-t-butyl-4-methylphenol and 3-t-butylchrysene. Single molecule calculations were compared with models for the molecular solids. Barriers were analyzed by decomposition into contributions from non-bonding interactions, delocalization, and relaxation of the molecular backbone.

Abstract
State-of-the-art molecular simulation technology remains largely restricted to pair-wise additive site-site Hamiltonians. Commercial stste-of-the-art programs with effective site-site pair potentials can deliver “animation”, but are of limited value as a predictive research tool. Water cannot be accurately represented by a pair-wise Hamiltonian. The same applies to the energy surfaces in macromolecules, polymers and network solids. Carbon is non-additive up to at least order 4. If many-body potentials of order n>2 or 3 are included in conventional MD, simulation of N-sites becomes N to power n times slower! To overcome this, we have developed a new approach. The method uses multidimensional arrays, for saving each site force (or potential in the case of MC) in a memory bank, as a function of its neighbourhood, rather like a neural net. Once “trained”, the many-dimensional array is state-independent and useful for wider N,V,T space. An algorithm has been developed for polarizable ions. The generating function for determining the “net” needn’t be the full Hamiltonian of the system. It needs only to get the n-body spatial distributions accurate up to the highest many-body term in the full Hamiltonian. For pair-wise systems, only the 2-body distribution is required. A rigorous perturbation expansion can then obtain the requisite properties from the configurations. The full Hamiltonian divides into a reference plus the perturbation. The generating function may be the whole reference part, but it will be shownthat it can be further truncated to include only the essential near-neighbor forces that determine the structural distributions. The advantage here is in the speed of simulation; it is independent of the complexity of the Hamiltonian or the degree of non-additivity. This method will be illustrated with ionic liquids, but extends to all non-additive molecular simulations of condensed materials.

Abstract
Zinc-containing matrix metalloproteinases (MMPs) are important drug targets in many inflammatory, malignant and degenerative diseases. Force field based methods such as the LIE approach and MM/PB/SA have been employed to calculate binding affinities of MMPs inhibitors. In such calculations, a bonded or nonbonded model has been adopted for the zinc ion, and the results are strongly affected by this ad hoc assumption. In addition, the calculations can't take into account the charge transfer and proton transfer in the formation of complexes.We now report a calculation of binding affinities for a series of stromelysin-1 (MMP-3) inhibitors using the linear scaling semi-empirical quantum chemistry method MOZYME. The inhibitors contain different zinc binding groups like carboxylic acid and hydroxamate. In this calculation, the whole protein and protein-ligand complexes were treated by a PM5 Hamitonian and the bonding characteristics of the zinc center and the charge / protonation states are automatically determined by the quantum mechanical wavefunction. Our results demonstrate that a combination of this QM method with the COSMO calculated solvation energy is a promising approach for calculating binding affinities in zinc-containing enzymes.

Abstract
Implicit solvent models, such as Generalized Born (GB) theory, have become popular alternatives to explicit solvation due to their improved computational efficiencies. However, implicit solvent models may not be sufficiently accurate for certain applications, such as free energy calculations. We present a novel method which combines explicit and implicit solvent models to obtain a balance between accuracy and computational efficiency. This method involves encapsulating a solute with a layer of water molecules and using GB theory to properly account for the reaction field around the irregular boundary of the system. Furthermore, we incorporate a multigrid algorithm to significantly speed up the electrostatic and GB pairwise interactions. Thanks to multigrid enhancements and the reduction of the number of explicit water molecules, our procedure is considerably faster than the conventional particle mesh Ewald method for the systems we have looked at so far. In an initial application, we used our method to assess the accuracy of different types of Poisson implicit solvation models. Specifically, we calculated the electrostatic charging free energies of various fixed conformations of two proteins. In our assessment of Poisson solvation models, we evaluated the relative merits of modifying van der Waals radii, varying the probe radius, and using various dielectric boundary definitions.

Abstract
A novel automated quantitative structure-activity relationship (QSAR) method has been developed based on the local linear regression approach. Activities of test set compounds are predicted as the weighted averaged activity of their nearest neighbors in the training set. The neighbors and their weights are defined by the Gaussian function, and its smoothing parameter is optimized during model building. ALL-QSAR method was applied to three datasets, including 48 dopamine D1-receptor antagonists and 48 anticonvulsants with known IC₅₀ values, and 250 phenolic compounds with known toxicities IGC₅₀. Models with R²=0.8¨C0.9 were built. Y-randomization tests showed no overfitting. These models were significantly better than those built for the same datasets using kNN, SVM, and PLS approaches. ALL-QSAR is a fast and reliable method to develop statistically robust and predictive models. They can be applied to screen chemical databases and virtual libraries to discover novel biologically active compounds.

Abstract
We present an automated software application that creates Bayesian neural network models as an aid to predict ADME/Tox properties. In-house models exist for several properties using company-wide or global data. In addition to these tools, it is interesting for researchers to make predictive models for data in smaller, more focused (local) datasets using similar modelling methodologies. A user submits a dataset of compounds for modelling and chooses several modelling parameters. Data pre-processing determines the feasibility of modelling the dataset. If data is adequate, then molecular descriptors are calculated and filtered, and data is passed to the modelling routines. Models are created using Flexible Bayesian Modelling (FBM) programs with an automatic relevance determination (ARD) factor. After a stable model is produced, irrelevant descriptors are removed from consideration in the next round of model training, and training continues until stopping criteria are met. The final list of models is searched for the highest quality factor (QF), and the best model is chosen for final training and subsequent validation. Confidence in validation set predictions can be assessed by distance-to-model measurements as well as analysis of the distribution of network errors in the training set.

Abstract
Formulation datasets typically contain relatively small amounts of data, but nonetheless artificial intelligence methods like neurofuzzy logic can be used to develop useful models that highlight the most important cause-and-effect interactions affecting end-use properties. In this paper we discuss some of the issues regarding model complexity as a function of the amount of data that is available, balancing the risk of overtraining with the desire to extract the maximum amount of information available. In particular, it is crucial that the correct statistical model selection criterion be used. Two examples, of pharmaceutical formulations, are used to illustrate our findings.

Abstract
FlexX has been among the first commercially available docking programs and established it's position as a leading software for structure-based design. Over the years a multitude of novel technology was added to the base code for incremental growth of a ligand within the active site of the protein. There is the rapid processing of combinatorial libraries, the consideration of water molecules, the ability to simultaneously handle ensembles of active site conformations and the guided search with user-defined pharmacophoric constraints to mention just a few. The latest developments on the technical end of things, besides a substantial speed-up due to code optimization, are the incorporation of smarts-based substructure recognition, facilitating rapid filters, the consideration of multiple protonation states and the exchange of bioisosteric groups. Python has been added as an interface to a generic multi-purpose programming language, allowing now also the FlexX-batch-processing. We will summarize the current status of the FlexX-docking software and the latest developments from currently running projects.

Special Equipment Needs: LCD projector for laptop

Abstract
Potency prediction of docked ligands depends not only on scoring but also on reliable docked geometries. Two different algorithms for thorough docking will be described. The first (MCDOCK) uses mutiple cycles of fast grid-box screened Monte Carlo searching. The search starts broadly and becomes progressively focused. The second method (ZIPDOCK) is near systematic. It uses a conformer compression algorithm to allow millions of docking poses to be screened in less than a minute. Both methods output multiple poses optimized in the binding site (full cartesian). An important feature is that selected parts of the site can move. Multiple docking poses were used to derive a new scoring function (CONTACT). Choice of docking method and number of rotatable bonds are found to have critical effects on the scores. Combining both docking methods significantly improves scores for flexible ligands.

Abstract
Docking is often described as consisting of two major components: a scoring function and a search method. Implicit in this is the representation of the molecules being docked, and how new candidate dockings are generated. Recent developments in AutoDock in each of these areas will be presented, including the development of a new empirical free energy scoring function, new search and optimisation methods, and a novel representation of molecular flexibility able to incorporate motion of both domains and side-chains in proteins.

Thanks to on-going advances in computer hardware, AutoDock is able to benefit from a bold, new way to interact with computational chemistry code, by using an extensible, object-oriented interpreter.

Progress in ease-of-use of AutoDock, and applications in protein-protein docking, covalent docking, and in silico high-throughput screening (HTS), will also be presented.

Abstract
Molecular docking is widely used to screen large compound collections for novel lead molecules that complement a receptor of known structure. Docking energy functions are approximate and many degrees of freedom are under-sampled. To understand where algorithms can be improved, we have turned to model systems where predictions can be tested in detail. We are using both highly simplified, cavity sites in T4 lysozyme, slightly more complicted cavity-like sites that are also open to solvent at one end, and full "drug-like" binding sites, the latter in b-lactamase. Predicted ligands are being tested for binding, geometry, and protein motion using x-ray crystallography. We hope to use this cycle of theory development and testing in a range of simple and more complicated sites to understand some of the weaknesses in our current docking algorithms, and to improve them.

Abstract
We present recent improvements to the LIGANDFIT program including algorithmic enhancements to improve positional and orientational sampling of the ligand and better selection of poses retained for scoring. Improved sampling is achieved by employing “site partitioning” where the binding site is further partitioned into smaller sites of various sizes and the ligand aligned to various partitioned sites by shape comparison. This procedure has been recently further refined by considering sites obtained by fusing various adjoining sites. In cases where the defined binding site is much larger than the size of the candidate ligand, the site partitioning technique significantly improves the quality of the docking. Results obtained with various Protein-Ligand complexes using this improved sampling will be presented. A technique for retaining poses that improves the overall diversity of the pose list will be discussed. Finally, ongoing work in the area of scoring function development will also be presented.

Abstract
Experimental proof is provided that sampling of low energy conformers is insufficient to reproduce protein-ligand binding geometries. eHiTS explores the conformational search space exhaustively, producing accurate docking poses at competitive speed. The customizable scoring function of eHiTS combines novel terms with traditional empirical and statistical approaches. Automatic training tools can adjust the scoring system to any set of experimental data. The program recognizes if the input receptor matches one of the protein families from its knowledge base and uses the appropriate scoring scheme that was trained for that specific family. Validation results of eHiTS are presented on a set of 50 PDB structures representing various DHFR-ligand complexes to demonstrate the ability of the program to accurately reproduce known binding poses. Cross docking results and enrichment results from a diverse library of 5000 ligands will also be presented to evaluate the selectivity of the scoring function. More information: http://www.simbiosys.ca/

Abstract
Bonding ideas, good and bad, emerge out of looking at calculations. Some which come from years of following Fritz Schaefer's work, will be described. More generally, we are approaching a time when ideas of bonding may emerge as much from mining theoretical data as experimental findings. With some interesting attendant tactical problems.

Abstract
Cycloaromatization reactions transform closed shell molecules into reactive diradical species – the process which has been used by nature to develop lethal chemical warfare of astounding power. Photochemical triggering of such processes may provide the temporal and spatial control needed to harness the record-breaking DNA-damaging power of these processes in the design of tumor-selective anticancer agents.

In this talk, I will give an example of how fundamental concepts of physical organic chemistry can be combined with computational and experimental studies to provide a better insight into the nature of factors controlling the efficiency of cycloaromatization reactions and lead to the discovery of new reactions with increased DNA-cleaving potential. In particular, I will discuss how the large effects of benzannelation and remote substituents on radical-anionic cyclizations originate from crossing of out-of-plane and in-plane MOs in the vicinity of transition states. This crossing leads to restoration of the aromaticity lost upon one-electron reduction of benzannelated enediynes. The trade-off between reduction potentials and cyclization efficiency as well as the possibilities of switching of enediyne cyclization modes (exo- or C1-C5 vs. endo- or C1-C6)) under kinetic or thermodynamic control conditions will also be outlined.

Abstract
Several research projects are highlighted involving our continuing pursuit of purely ab initio methods for thermochemical accuracy to the level of 0.1 kcal/mol. Molecular principal and partial wave expansions of problematic systems have been computed for both conventional and R12 correlation methods, with one-particle basis sets extending to k spherical harmonics, thus probing fundamental accuracy limits and demonstrating the superior convergence behavior of explicitly correlated methods. New cusp-satisfying ansätze for pair correlation functions are investigated by means of analytic work on atomic models. The problem of accurately computing connected quadruple excitation effects on bond dissociation energies is elucidated by benchmark full CCSDTQ studies. Improved formulas are derived for computing anharmonic zero-point vibrational energies from quartic force fields, with application to species as large as the amino acid proline. Finally, chemical applications are reported ranging from soot formation intermediates to proton affinity scales for biomolecules.

Abstract
One of the most intriguing effects the fundamental weak interaction may have in chemistry is the parity violating energy difference (ΔE_pv) between two mirror-image molecules. While a successful measurement of ΔE_pv is still lacking, considerable progress has been made in the past few years in the theoretical prediction of molecular parity violating effects. The methods developed for this purpose include correlated one-component linear response approaches (MCLR) and four-component Dirac-Hartree-Fock (DHF) schemes.

Here we present a two-component approach to molecular parity violation which employs the zeroth order regular approximation (ZORA) and allows to combine the strengths of previous one-component and four-component schemes since spin-orbit coupling is treated self-consistently. For benchmark systems of the type H₂X₂ (X=chalcogen) we compare results obtained with the ZORA approach with those of former MCLR and DHF treatments and we present studies on systems of significance for a spectroscopic proof of molecular parity violation.

Abstract
We report computational and experimental investigations aimed at gaining an understanding of the energetic and structural properties of chemically functionalized single-walled carbon nanotubes. The main focus is on fluorinated single-walled carbon nanotubes, for which we report a detailed investigation of the fluorine loss and an assessment of their propensity to form CF-HO hydrogen bonds. More generally, detailed density functional computations on finite carbon nanotube clusters of increasing size and periodic systems with increasingly larger unit cells enable an evaluation of finite lengths effects and ultimately the determination of the binding energies of addends (fluorine and carbenes) to the nanotube sidewalls depending on chirality and diameter. This approach also allows comparing the reactivity to sidewalls with defects.

Abstract
Substituent and strain effects that weaken interallylic bonding are predicted to be capable of transforming the boat transition structure for the Cope rearrangement of semibullvalene into the global energy minimum. B3LYP and ab initio calculations have been used to compute the strengths of the interallylic bonds in semibullvalenes that are predicted to have delocalized equilibrium geometries. The insights that these calculations provide into the question of whether such semibullvalenes are "bishomoaromatic" will be discussed.

Abstract
When designing new molecules to fit a target binding site, it is easy to see where hydrophobic or hydrogen bonding ligand atoms should lie. However, it is often very difficult to find a chemical fragment that will place all these atoms into optimal positions. To be successful, a de novo program should have a rich repertoire of diverse motifs generated rapidly and with accurate geometry. The de novo program, AlleGrow (1), has achieved these goals and has been shown to generate ligands with geometries similar to those found in x-ray structures. Because AlleGrow lacked the ability to create the large number of polycyclic structures found in many drugs, a library of ~5000 heterocycles has been added. AlleGrow explorations of the binding sites of thermolysin, Src SH2 and CDK2 will be reported.

(1) AlleGrow is a second generation program based on GrowMol (R.S. Bohacek, C. McMartin, JACS (1994) 116, 556—5571).

Abstract
Numerous products are available to discover viable drug leads through virtual screening. However, drug refinement entails a collaborative effort between computational and synthetic chemists. Ideal design tools should empower chemists to utilize their knowledge of the active site and exploit their synthetic intuition. The MindRocket is a ligand development system that allows chemists to rapidly generate, dock, visualize, and iteratively amend novel chemical structures on the fly. Optimal poses are quickly determined with full ligand-receptor flexibility using scoring functions derived from user structure activity data. Immediate feedback is provided for refinement. Series of structures can be enumerated using scaffolds via Markush-like descriptors. A key aspect of this software is novel technology incorporating user constraints and ADMET properties to govern synthetic feasibility and rapidly eliminate undesired chemical constructs. This presentation will discuss the MindRocket’s unique capabilities along with specific applications that demonstrate its value in ligand optimization.

Abstract
De novo ligand design systems have undergone substantial development over the past decade, to the extent that several of the currently available systems are capable of suggesting large numbers of ligands with high estimated affinity for the target protein. This has led to increased emphasis on the development of methods for the automated selection of specific ligands for synthesis. Clearly estimation of synthetic accessibility must be a key component of any ligand scoring system. In the SPROUT system, this problem has been addressed in three ways: a) the stand alone CAESA program for estimation of synthetic accessibility; b) the SynSPROUT program in which the ligand construction process mimics reactions taken from a knowledge base; c) a new feature in SPROUT which assesses synthetic complexity by a multilevel comparison of generated structures with structures from databases of previously synthesised potential drugs and also from supplier catalogs.

Abstract
Fundamental issues with de novo modifications of lead compounds include the presentation of results to chemists, incorporation of chemical sensibility, ADME-like properties, and other preferred properties, as well as the all-important prediction of affinity. We have integrated the de novo design engine, LUDI (Böhm, JCAMD 6:61 1992; Böhm, JCAMD 6:593 1992), with C2.Minimizer, C2.QSAR+, and C2.Descriptor+ to produce, in an automated fashion, either (1) novel inhibitors that are selected as survivors from improving generations of inhibitors, or (2) combinatorial enumeration of all derivatives of a given scaffold that satisfy user-selected rules. Both modes will be presented with validations and correlations to published experimental data.

Abstract
Recent progress will be presented in the use of optimization approaches to study and design binding partners for proteins. The focus will be on methods for simultaneously satisfying packing and electrostatic constraints in a computationally efficient manner. Illustrative examples involving lead optimization as well as lead discovery will be presented, as well as cases of improving affinity and strategies for altering specificity of binding interactions.

Abstract
Compton scattering, (e,2e) spectra, and PES all give some information about the momentum dependence of orbitals. In this lecture we will compare some results using large CI wave functions and DFT. Even though DFT does not give a momentum density, use of Kohn-Sham orbitals as though they formed a wave function does yield reasonable results in momentum space. The average kinetic energy, however, is the Kohn-Sham kinetic energy so clearly this momentum distribution is not exact.

Abstract
"Aromaticity" also is applicable to inorganic compounds and clusters, although its detailed understanding is even more complicated. Thus, both the degree of bond length equalization and simple electron-counting rules often fail to characterize aromaticity satisfactorily. (1) The recently claimed 4p antiaromatic Al4Li3- cluster with alternating bond lengths is actually aromatic since the s-aromaticity dominates its p-antiaromaticity. (2) Isoelectronic molecules may behave quite differently; highly symmetrical (Td, Oh, Ih etc.) clusters with equal bond lengths may either be aromatic or antiaromatic, depending on the elements and the substituents involved. Aromatic stabilization energies strongly depend on the reference molecules, and are difficult to evaluate. In contrast, NICS (Nucleus-Independent Chemical Shift), a magnetic measure of aromaticity, not only is simple and effective, but also provides useful prediction of stable species, and deep insights into the seemingly erratic behavior of inorganic compounds and clusters.

Abstract
The diversity of structure and reactivity associated with carbenes, nitrenes, arynes, nitrenium ions, and related reactive intermediates has long been a source of fascination for organic chemists, and of consternation for computational chemists. Accurate predictions of both singlet- and triplet-state electronic structures can be particularly challenging owing to the typically multiconfigurational nature of at least one of the relevant wave functions. The quantitative and qualitative lessons learned from organic systems may be applied to inorganic systems with similar frontier orbital characteristics. This can become an issue when single-determinantal density functional theory, usually well suited to the description of many properties of 1:1 metal:dioxygen complexes, fails to accurately compute state-energy splittings because of limitations associated with diradical character. A key example of this latter situation will be discussed within the context of the copper-dioxygen species of Tolman and co-workers; general historical and algorithmic discussion will sample from the organic diradicals too.

Abstract
The reliable prediction of specific rotation in chiral molecules is a long-standing goal of chemistry. Although empirical models such as the quadrant and octant rules have been used for decades, they provide only qualitative results, at best. We have recently developed a new series of programs for the computation of frequency-dependent optical rotation using coupled cluster linear-response theory. In this work, we consider the importance of basis set size and character as well as high-level electron correlation effects on coupled cluster optical rotation angles for rigid species such as S-methyloxirane. We find that although both basis-set and electron-correlation contributions are significant, the former are paramount: qualitatively incorrect results can be obtained even for relatively large basis sets. Furthermore, we find an unexpected frequency dependence: errors relative to experiment are sometimes larger for shorter wavelengths than for the sodium D-line.

Abstract
Hydron transfers, particularly of protons, are perhaps the most common reactions in nature and their mechanisms are of fundamental importance in chemistry. This paper will discuss recent hybrid and quantum mechanical computational studies which have been employed to understand the novel catalytic tunnelling mechanisms now evident in enzymes and to model experimental intramolecular mimics in the aqueous phase.

Abstract

Abstract
Electronic structure calculations have become capable of predicting a large number of rovibrational band origins and other spectroscopic properties to within a wavenumber or better. Such state-of-the-art ab initio electronic structure computations, resulting in a highly accurate potential energy (PES) and dipole moment (DMS) surface for the prototypical triatomic molecule H2O, are reviewed highlighting the hierarchy of the physical effects to be considered. The use of high-order force fields for the representation of the PES is also discussed. Different variational strategies for solving the nuclear motion problem are discussed next. Emphasis is put either on the simplicity of the approach, provided by the discrete variable representation (DVR) of the Hamiltonian, or on the utility of the solution strategy in handling singularities in the Hamiltonian. Generalization of the DVR approach and its accuracy is addressed. Representative numerical results are presented for triatomic systems.

Abstract
Over the last few years, the amount of chemical and biological data generated during the drug discovery process has continued to grow rapidly. This data is quite heterogeneous and resides in a variety of formats - from local flat files to relational databases. Consequently, for bench scientists, it is often difficult - or even impossible - to search and retrieve relevant chemical and biological data for compounds and protein targets of interest. This poster presents a unique solution to this dilemma. DeltaSoft’s ChemCart and LION’s SRS Gateway for Oracle products have been integrated to provide scientists with an easy-to-use, customizable tool to search and retrieve structures, primary/secondary screening results, images, gene, protein, clinical, and expression data. Using the NCI gene expression data set, we provide an example of retrieving relevant chemical and biological information into to a single cohesive view.

Abstract
Molecular modeling provides a mechanistic hypothesis at the molecular level for the constitutive activation recently observed and reported for tyrosine protein kinases Flt-3 and c-Kit. Three-dimensional homology models for the active and inactive forms of these two kinases were made. Comparison of these models at the molecular level reveals that mutations of specific residues located in the activation loop (D835X and 836-deletion in Flt-3; D816V in c-Kit) as well as a 6-base pair insertion at residue 840 in Flt-3 operate in a similar way. Each mutation tends to weaken the forces that maintain the activation-loop folded inwards. None of the mutations are found to particularly stabilize the active state directly. The reason why the equilibrium is shifted towards the gate-open conformation of the protein is because the mutations, at least in these models, are found to critically destabilize the inactive conformational state of the kinase.