DDR

DRUG DEVELOPMENT RESEARCH 75 : 412–418 (2014)

Research Overview

Computational Approaches for Drug Discovery Che-Lun Hung1* and Chi-Chun Chen2,3 1 Department of Computer Science and Communication Engineering, Providence University, Taichung City, 43301, Taiwan 2 Bioinformatics Program, Taiwan International Graduate Program, Institute of Information Science, Academia Sinica, Taipei 11529, Taiwan 3 Institute of Bioinformatics and Structural Biology, National Tsing Hua University, Hsinchu 30013, Taiwan

Strategy, Management and Health Policy Enabling Technology, Genomics, Proteomics

Preclinical Research

Preclinical Development Toxicology, Formulation Drug Delivery, Pharmacokinetics

Clinical Development Phases I-III Regulatory, Quality, Manufacturing

Postmarketing Phase IV

ABSTRACT Cellular proteins are the mediators of multiple organism functions being involved in physiological mechanisms and disease. By discovering lead compounds that affect the function of target proteins, the target diseases or physiological mechanisms can be modulated. Based on knowledge of the ligand–receptor interaction, the chemical structures of leads can be modified to improve efficacy, selectivity and reduce side effects. One rational drug design technology, which enables drug discovery based on knowledge of target structures, functional properties and mechanisms, is computer-aided drug design (CADD). The application of CADD can be cost-effective using experiments to compare predicted and actual drug activity, the results from which can used iteratively to improve compound properties. The two major CADD-based approaches are structure-based drug design, where protein structures are required, and ligand-based drug design, where ligand and ligand activities can be used to design compounds interacting with the protein structure. Approaches in structure-based drug design include docking, de novo design, fragment-based drug discovery and structure-based pharmacophore modeling. Approaches in ligand-based drug design include quantitative structure–affinity relationship and pharmacophore modeling based on ligand properties. Based on whether the structure of the receptor and its interaction with the ligand are known, different design strategies can be seed. After lead compounds are generated, the rule of five can be used to assess whether these have drug-like properties. Several quality validation methods, such as cost function analysis, Fisher’s cross-validation analysis and goodness of hit test, can be used to estimate the metrics of different drug design strategies. To further improve CADD performance, multi-computers and graphics processing units may be applied to reduce costs. Drug Dev Res 75 : 412–418, 2014. © 2014 Wiley Periodicals, Inc. Key words: computers; drug design

INTRODUCTION

Proteins represent the basis of cellular and tissue function being responsible for multiple functions within organisms. By modulating target protein function in both normal and pathophysiology condition, it is possible to control tissue homeostasis. Small molecules can affect the function of target proteins to act as drugs and their chemical structures can be modified to improve their drug-like properties and reduce side effects and © 2014 Wiley Periodicals, Inc.

*Correspondence to: Che-Lun Hung, Department of Computer Science and Communication Engineering, Providence University, Taichung City, 43301, Taiwan. E-mail: [email protected] Published online in Wiley Online Library (wileyonlinelibrary .com). DOI: 10.1002/ddr.21222

COMPUTATIONAL APPROACHES FOR DRUG DISCOVERY

toxicities. Historically, many drugs have been discovered by serendipity; however, the development of new enabling technologies, including parallel chemical synthesis, high through screening and recombinant target expression, has led, to varying degrees depending on the specific target, to a more rational approach to drug design. Rational drug design provides the framework to design compounds with high specificity, potency, efficacy and favorable pharmacokinetics based on the knowledge of target structure, functional properties and mechanisms using computer-aided drug design (CADD). CADD allows the simulation of interactions between active compounds and their targets, including receptors, enzymes and transporters [Silverman and Holladay, 2014]. Based on the simulation results, researchers can design complementary structures for the targets, and then screen compound databases to identify compound “hits.” Initially CADD is utilized to computationally simulate active compounds, to screen compound databases, and to optimize lead compounds accelerating the drug discovery process and minimizing resource requirements. The actual activities of optimized leads can be tested in in vitro assays with the results being used to improve the simulation models. CADD approaches can be classified into two groups: structure-based drug design and ligand-based drug design.

413

Docking

structure with another molecule. The basic idea is consistent with the seminal lock and key approach to pharmacology, where a ligand functions as a key, that needs the correct relative orientation to interact with the protein surface, which functions as a lock [Jorgensen, 1991]. The first step of molecular docking is to generate a group of spheres that represent active sites on the target and use these to conceptualize a series of possible binding sites. Based on binding distance and energy between the ligand and binding sites, interaction models and affinities can be calculated to simulate optimal interactions and thus filter possible pharmacophores. Based on the molecular flexibility during docking, three types of docking models can be envisaged: (i) rigid docking, where the structures of ligand and target cannot change during docking; (ii) semi-flexible docking, where only the structure of the ligand changes during docking, but the structure of the target is rigid; and (iii) flexible docking, where both the ligand and target are treated as flexible structures. In the flexible docking, the complexity and cost of computational calculation is higher than the other two models [Mohan et al., 2005]. Ligands have distinctive pharmacophores and functional groups that can interact with drug targets to form complexes that initiate (agonists) or prevent (antagonists) functional changes in target function. Five molecular forces can affect binding affinity and energy to affect the interaction: (i) covalent bonds that share electron pairs between the ligand and target atoms— these are relatively strong and difficult to break with the number of electron pairs corresponding to the number of covalent bonds; (ii) van der Waals interactions, weak and nonspecific attractive forces caused by transient dipoles; (iii) hydrophobic interactions, the result of aggregation of nonpolar molecules or substances that cannot hydrogen bond with water molecules; (4) hydrogen bonds, the interaction between a hydrogen atom in a molecule with a partially positively charge and another atom with unpaired electrons; (5) ionic interactions, the attraction between a cation and an anion [Lodish, 2012]. Three representative docking software programs demonstrate different docking strategies:

Docking has been extensively applied to protein–DNA interactions [van Dijk et al., 2006], protein–protein interactions [Gray et al., 2003; Mintseris et al., 2005] and protein–ligand interactions [Morris et al., 1996, 1998; Rarey et al., 1996; Ewing and Kuntz, 1997]. In molecular simulation, docking [Perola et al., 2004; Leach et al., 2006] is utilized to predict the conformer of a molecule that can form a stable complex

(1) Dock [Ewing and Kuntz, 1997], which uses a geometric matching algorithm to simulate interaction between ligand and target to find the binding mode with the lowest energy. Dock also provides different scoring methods to estimate ligand–target interactions. (2) AutoDock [Morris et al., 1996, 1998], which uses simulation annealing and genetic algorithms to

STRUCTURE-BASED DRUG DESIGN

Due to the advances made in molecular and structural biology over the past 50 years, many more protein structures are available [Berman et al., 2000] allowing the application of homology modeling to predict protein structures that are unknown. When the target protein structure is known, structure-based approaches, such as docking, de novo ligand design, fragment-based drug discovery (FBDD) and structure-based pharmacophore modification can be used.

Drug Dev. Res.

414

HUNG AND CHEN

search for the best ligand–target interaction location and semi-empirically calculate ligand–target complex free energy. (3) FIexX [Rarey et al., 1996], which combines multiple drug design methods and applies flexible docking to perform ligand–receptor docking. De Novo Ligand Design De novo ligand design uses computational power to design ligands that have proper functional groups and conformer that can interact with a target. The main goal is to link separated functional groups that can interact with active sites in a simulated compound that fits with chemical principles. The simulated compound can serve as a template for database searching to find existing compounds or to develop synthetic methods to reduce the number of intermediates. When the protein–ligand complex structure is known, the initial chemical structure of a ligand can be changed by checking if the simulated compound can improve the interaction affinity on the same location. There are two strategies for de novo ligand design: (i) Grow: After adding a molecule as a seed into the target, link molecular fragments that fit to the target for the seed to complete the ligand structure with the best to the target; and (ii) Link: After adding molecular fragments with the best interaction energy with the target, the fragments are linked to form a new ligand structure. There are three representative de novo ligand design software packages: (1) LUDI [Bohm, 1992; Prathipati and Saxena, 2006], which considers hydrophobic interactions and hydrogen bonds during fragment and target interaction site generation. (2) GRID [Kastenholz et al., 2000], which classifies the active target sites into grids, and then adds probes to calculate the interaction energy between probes and grids to find the specific groups that can interact with the targets. Van der Waals interaction, hydrogen bonds and electrostatics are parts of the calculation. (3) MCSS (multiple copy simultaneous search) [Caflisch et al., 1993], which uses force fields to calculate the interactions between fragments and active sites to find chemical functional groups that can interact with the targey.

to a target and convert the fragments into lead compounds. The fragments need to fit the rule of three [Congreve et al., 2003]: (i) The molecular weight should be less than 300; (ii) The number of hydrogen bond donors and acceptors should not be more than 3 each; (iii) The calculated Log P (CLogP) should not exceed 3. The fragments are used as building blocks to create more complex leads with improved affinity. The basics of the strategy are as follows: (i) when a fragment has a low affinity, it may bind the target sufficiently to be optimized; (ii) a new chemical series/pharmacophore can be generated by linking fragments. Two FBDD software tools are as follows: (i) SPROUT [Law et al., 2003], which recognizes different types of active binding sites within a target, and then uses docking to search for fragments, which can interact with active binding sites, so that the selected fragments can be joined. The software provides a module to score and sort the output of leads based on their binding affinity; and (ii) GANDI [Dey and Caflisch, 2008], which uses a genetic algorithm to search fragments based on force field energy and known binding characteristics to optimize the selected fragments. Structure-based pharmacophore Functional groups can markedly influence compound activity with some changes in ligand structure dramatically affecting ligand–target interactions, while others do not. Compounds with similar activity often share the same features leading to a pharmacophore with features that contribute to activity. Thus the consensus pharmacophore has spatial patterns of abstract features responsible for its bioactivity. The main concept of structure-based pharmacophore design is to use docking for calculating target–ligand interactions to generate structure-based pharmacophore models, A representative structure-based pharmacophore software program is e-Pharmacophore [Loving et al., 2009; Salam et al., 2009], which applies a docking software, Glide, with an extra precision mode, to calculate protein–ligand interactions with excluded volumes being used for the calculation. It has two modes, singlemode, where input is a protein-ligand complex or protein and a known binding ligand, and fragmentmode, where fragments are used to find energetically compatible binding sites to generate models. LIGAND-BASED DRUG DESIGN

Fragment-Based Drug Discovery (FBDD) The basic concept of FBDD [Erlanson et al., 2004] is to identify small chemical fragments that bind Drug Dev. Res.

In ligand-based design a group of active molecules is used to infer active ligand structure to design and optimize compounds. It can be separated into two

COMPUTATIONAL APPROACHES FOR DRUG DISCOVERY

classes, quantitative structure–affinity relationships (QSARs) and pharmacophore-based design. QSAR Three-dimensional QSAR (3D-QSAR) is a major aspect of QSAR, where the chemical features of active ligands can be assembled to form a specific mode of active ligands for the target active sites in 3D space. Hydrogen bonds, charge interactions and hydrophobic areas are included in chemical features in 3D-QSAR. The outcome model sets the conditions for ligand– macromolecule interactions and can be used to search databases to discover lead compounds. There are two 3D-QSAR software packages: (1) CoMFA [Cramer et al., 1988; Labrie et al., 2006; Awale and Mohan, 2008] with the basic approach that if a group of similar ligands modulate a given target in the same way, their bioactivity should depend on shared molecular fields. There are three steps in the calculation: (i) finding similar structures between different ligands, and then assembling these; (ii) defining a cuboid around the assembled ligand structures that is separated into grids, and probed to analyze steric-electrostatic or hydrophobic-hydrogen fields in each grid; (iii) using partial least-squares analysis to create relationships between ligand activity and the molecular field features. (2) CoMSIA [Klebe et al., 1994; Bohm et al., 1999; Labrie et al., 2006], which is similar to CoMFA. However, in the third step, CoMSIA is stricter than CoMFA, where CoMSIA considers steric, hydrophobic, electrostatic, acceptor and donor separately, and uses Gaussian function during the calculation. The strategy helps CoMSIA to provide more stable results and avoid some defects caused by CoMFA.

415

Hypotheses can be created using information from x-ray or nuclear magnetic resonance (NMR) activity with structure relationship analysis results based on a training set being used to postulate pharmacophores based on chemical characteristics. Two bases of this strategy are as follows: (i) All compounds in a training set have the same binding pattern with the same targets; and (ii) Compounds with more binding interactions should have higher activity compared with those having less binding actions. The selection of training set determines if software can successfully generate hypotheses. There are four principles for training set selection: (i) It should have enough compounds to guarantee that the generated pharmacophores are statistically significant; (ii) Compounds in the training set should have different structures, and their activity ranges should be 4–5 orders of magnitude; (iii) Selected compounds should have clear, nonredundant and unbiased information; (iv) The compound with the highest activity should be in the training set since it can provide the most important effect during pharmacophore generation. GASP is a pharmacophore software [Jones et al., 1995] that uses a genetic algorithm to flexibly map compounds to find the best pharmacophore. Two characters of GASP use a genetic algorithm with unique fitness function to optimize pharmacophore models. Besides software for individual strategies, there are software programs like PHASE that combine both 3D-QASR and pharmacophore approaches. PHASE [Dixon et al., 2006a, 2006b] identifies functional groups in high affinity ligands that are spatially arranged and are essential for ligand bioactivity.

APPLICATIONS OF CADD

Pharmacophore

Based on target and ligand, there are four options, known ligand, unknown ligand, known target and unknown target, that can be used to select the CADD strategy to be used.

The main concept of a pharmacophore-based approach is to find a series of common pharmacophores within ligands, where each pharmacophore includes specific chemical bonds, acceptors, donors, cations, anions, and hydrophobic centroids that are treated as ligand–target interaction active sites. Pharmacophore models create validatable hypotheses that involve the spatial positions of functional groups that interact spatially. Each feature is conceptualized as being surrounded by a sphere representing a space occupied by an atom or several atoms. In synthetic molecules, different structures often have similar bioactivity.

(1) When both the ligand and target are known, docking and structure-based pharmacophores can be used to design compounds. In the docking case, a ligand database can be regarded as input data to calculate binding energy and orientation. In the case of a structure-based pharmacophore, the output models can be regarded as references to search ligand database and for chemical synthesis. (2) When only the target is known, de novo ligand design and FBDD can be applied to facilitate compound design. Fragments can be chosen based on active sites, and linked as a new compound structure that adheres to chemical principles. Drug Dev. Res.

416

HUNG AND CHEN

(3) When only ligands are known, 3D-QSAR and pharmacophore can be used to facilitate compound design. The main idea of 3D-QSAR is to align a series of compounds with identical structures that have different activities. The pharmacophore is used to analyze compounds with shared bioactivity but not structure. (4) When both of ligand and target are unknown, there is no effective drug design strategy. After generating lead compounds, Lipinski’s rule of five [Lipinski et al., 1997], can be used to evaluate if the leads have drug-like properties. The rule of five states that: (1) The number of hydrogen bond donors should be less than 5. (2) The number of hydrogen bond acceptors should be less than 10. (3) The molecular weight should be less than 500. (4) CLogP should be less than 5, or theMoriguchi Log P (MlogP) should be less than 4.15. (5) When the compound classes are biological transporter substrates, the rule of five does not apply. QUALITY VALIDATION

There are three methods for quality validation of models: cost function analysis, Fisher’s cross-validation analysis and goodness of hit (GH) test.

pound activities are randomly reassigned. (ii) Based on an original and a new training set, their total cost values and correlation values are calculated, respectively. The outcome values indicate the relationship between predicted activities and actual activities. (iii) The difference between the original and a new training set are compared. If the original training set has a higher correlation value and a lower total cost value, the model is considered to be improved. GH (Güner–Henry) Test This test is used to check if the result of screening compound databases has statistical significance. When a number of known compounds are active, the test can be applied to validate the model. The equation of the GH test is

H − Ha ⎞ ⎛ Ha (3 A + Ht ) ⎞ ⎛ GH test score = ⎜ × 1− t ⎟ ⎝ ⎝ ⎠ D−A ⎠ 4Ht A where D is the size of the database, A is the number of active compounds in the database, Ht is the number of predicted active compounds, and Ha is the number of active compounds within the predicted active compounds. The GH test score of a good model should not be less than 0.5. Future Work

Cost Function Analysis The cost function can be used to analyze model complexity and bias based on chemical characteristics and weighting and the difference between predicted and actual results. Each hypothesis has a total cost value. When the cost is smaller the model is better. Two cost values, fixed cost, where predicted results and actual results are identical, and null cost, where predicted results are generated by a pharmacophore which has no features, and are then used to analyze the quality of each hypothesis. If the cost value of a hypothesis is closer to the fixed cost and it is smaller than the null cost, the hypothesis is considered to be better. There is also a configuration cost that is calculated based on space complexity. A good model should have a configuration cost less than 17. Fisher’s Cross-Validation Analysis Fisher’s cross-validation analysis is used to check if the predicted activities and actual activities are correlated. This method consists of three steps: (i) ComDrug Dev. Res.

Advances in computer science, molecular and structural biology, combinatorial chemistry, quantum chemistry, pharmaceutical chemistry and molecular dynamics in the last decade, have allowed the more widespread, cost-effective use of CADD to enable drug discovery research. While multi-computer approaches, e.g., workstation clusters, personal computer clusters, massive parallel processors (MPP) and grid computing, can facilitate CADD, only large research groups or national research centers can afford large-scale multi-computer system although cloud computer services [Dalpé and Joly, 2014]. General-purpose graphics processing units (GPGPUs) have been applied to solve computationintensive problems of a variety of scientific domains [Guerrero et al., 2014]. GPGPUs have grown from a few cores to over a hundred and has significantly enhanced computational power. In 2006, nVIDIA proposed the Compute Unified Device Architecture (CUDA https://developer.nvidia.com/about-cuda) involving Single Instruction Multiple Threads computing architecture [SIMT; Liu et al., 2013] that allows

COMPUTATIONAL APPROACHES FOR DRUG DISCOVERY

417

computing thread to execute instructions, facilitating decision-based execution that is not provided by the more common-model, single instruction multiple data (SIMD). Both multi-computers and GPGPUs can significantly improve the CADD performance. Therefore, it is crucial to apply new high performance computing technology to improve drug design based approaches in the drug discovery process.

Ewing TJA, Kuntz ID. 1997. Critical evaluation of search algorithms for automated molecular docking and database screening. J Comput Chem 18:1175–1189.

ACKNOWLEDGMENT

Jones G, Willett P, Glen RC. 1995. A genetic algorithm for flexible molecular overlay and pharmacophore elucidation. J Comput Aided Mol Des 9:532–549.

This research was partially supported by the National Science Council under Grants 100–2221-E126-007-MY3.

Gray JJ, Moughon S, Wang C, Schueler-Furman O, Kuhlman B, Rohl CA, Baker D. 2003. Protein-protein docking with simultaneous optimization of rigid-body displacement and side-chain conformations. J Mol Biol 331:281–299. Guerrero GD, Imbernón B, Pérez-Sánchez H, Sanz F, García JM, Cecilia JM. 2014. A performance/cost evaluation for a GPU-based drug discovery application on volunteer computing. Biomed Res Int 2014. 474219.

Jorgensen WL. 1991. Rusting of the lock and key model for protein-ligand binding. Science 254:954–955.

REFERENCES

Kastenholz MA, Pastor M, Cruciani G, Haaksma EE, Fox T. 2000. GRID/CPCA: a new computational tool to design selective ligands. J Med Chem 43:3033–3044.

Awale M, Mohan CG. 2008. 3D-QSAR CoMFA analysis of C5 substituted pyrrolotriazines as HER2 (ErbB2) inhibitors. J Mol Graph Model 26:1169–1178.

Klebe G, Abraham U, Mietzner T. 1994. Molecular similarity indices in a comparative analysis (CoMSIA) of drug molecules to correlate and predict their biological activity. J Med Chem 37:4130–4146.

Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig H, Shindyalov IN, Bourne PE. 2000. The protein data bank. Nucleic Acids Res 28:235–242.

Labrie P, Maddaford SP, Fortin S, Rakhit S, Kotra LP, Gaudreault RC. 2006. A comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) of anthranilamide derivatives that are multidrug resistance modulators. J Med Chem 49:7646–7660.

Bohm HJ. 1992. The computer program LUDI: a new method for the de novo design of enzyme inhibitors. J Comput Aided Mol Des 6:61–78. Bohm M, St rzebecher J, Klebe G. 1999. Three-dimensional quantitative structure-activity relationship analyses using comparative molecular field analysis and comparative molecular similarity indices analysis to elucidate selectivity differences of inhibitors binding to trypsin, thrombin, and factor Xa. J Med Chem 42:458– 477. Caflisch A, Miranker A, Karplus M. 1993. Multiple copy simultaneous search and construction of ligands in binding sites: application to inhibitors of HIV-1 aspartic proteinase. J Med Chem 36:2142– 2167. Congreve M, Carr R, Murray C, Jhoti H. 2003. A “rule of three” for fragment-based lead discovery? Drug Discov Today 8:876–877. Cramer RD, Patterson DE, Bunce JD. 1988. Comparative molecular field analysis (CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. J Am Chem Soc 110:5959–5967.

Law JMS, Fung DYK, Zsoldos Z, Simon A, Szabo Z, Csizmadia IG, Johnson AP. 2003. Validation of the SPROUT de novo design program. Journal of Molecular Structure: THEOCHEM 666:651– 657. Leach AR, Shoichet BK, Peishoff CE. 2006. Prediction of proteinligand interactions. Docking and scoring: successes and gaps. J Med Chem 49:5851–5855. Lipinski CA, Lombardo F, Dominy BW, Feeney PJ. 1997. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv Drug Deliv Rev 23:3–25. Liu Y, Wirawan A, Schmidt B. 2013. CUDASW++ 3.0: accelerating Smith-Waterman protein database search by coupling CPU and GPU SIMD instructions. BMC Bioinformatics 14:117. Lodish HF. 2012. Molecular cell biology. New York, W.H. Freeman.

Dalpé G, Joly Y. 2014. Opportunities and challenges provided by cloud repositories for bioinformatics enabled drug discovery. Drug Dev Res doi: 10.1002/ddr.21211.

Loving K, Salam NK, Sherman W. 2009. Energetic analysis of fragment docking and application to structure-based pharmacophore hypothesis generation. J Comput Aided Mol Des 23:541–554.

Dey F, Caflisch A. 2008. Fragment-based de novo ligand design by multiobjective evolutionary optimization. J Chem Inf Model 48:679–690.

Mintseris J, Wiehe K, Pierce B, Anderson R, Chen R, Janin J, Weng Z. 2005. Protein-Protein Docking Benchmark 2.0: an update. Proteins 60:214–216.

Dixon SL, Smondyrev AM, Knoll EH, Rao SN, Shaw DE, Friesner RA. 2006a. PHASE: a new engine for pharmacophore perception, 3D QSAR model development, and 3D database screening: 1. Methodology and preliminary results. J Comput Aided Mol Des 20:647–671.

Mohan V, Gibbs AC, Cummings MD, Jaeger EP, DesJarlais RL. 2005. Docking: successes and challenges. Curr Pharm Des 11:323–333.

Dixon SL, Smondyrev AM, Rao SN. 2006b. PHASE: a novel approach to pharmacophore modeling and 3D database searching. Chem Biol Drug Des 67:370–372. Erlanson DA, McDowell RS, O’Brien T. 2004. Fragment-based drug discovery. J Med Chem 47:3463–3482.

Morris GM, Goodsell DS, Huey R, Olson AJ. 1996. Distributed automated docking of flexible ligands to proteins: parallel applications of AutoDock 2.4. J Comput Aided Mol Des 10:293–304. Morris GM, Goodsell DS, Halliday RS, Huey R, Hart WE, Belew RK, Olson AJ. 1998. Automated docking using a Lamarckian genetic algorithm and an empirical binding free energy function. J Comput Chem 19:1639–1662.

Drug Dev. Res.

418

HUNG AND CHEN

Perola E, Walters WP, Charifson PS. 2004. A detailed comparison of current docking and scoring methods on systems of pharmaceutical relevance. Proteins 56:235–249.

Salam NK, Nuti R, Sherman W. 2009. Novel method for generating structure-based pharmacophores using energetic analysis. J Chem Inf Model 49:2356–2368.

Prathipati P, Saxena AK. 2006. Evaluation of binary QSAR models derived from LUDI and MOE scoring functions for structure based virtual screening. J Chem Inf Model 46:39–51.

Silverman RB, Holladay MW. 2014. The organic chemistry of drug design and drug action, 3rd ed. San Diego, CA, Academic Press, p 123–202.

Rarey M, Kramer B, Lengauer T, Klebe G. 1996. A fast flexible docking method using an incremental construction algorithm. J Mol Biol 261:470–489.

van Dijk M, van Dijk AD, Hsu V, Boelens R, Bonvin AM. 2006. Information-driven protein-DNA docking using HADDOCK: it is a matter of flexibility. Nucleic Acids Res 34:3317–3325.

Drug Dev. Res.

Computational approaches for drug discovery.

Cellular proteins are the mediators of multiple organism functions being involved in physiological mechanisms and disease. By discovering lead compoun...
77KB Sizes 0 Downloads 10 Views