The process of bringing new and innovative drugs, from conception and synthesis through to been approved on the market can take the pharmaceutical industry 8–15 years, and cost approximately $1.8 billion (the average cost associated with this process has been estimated to be $900 million) (1, 2).
Just for the clinical period (i.e., clinical trial and long-term animal testing), the estimated costs per approved new chemical entities depend on average out-of-pocket clinical phase costs, attrition rates across phases, the probability of marketing approval, and development and regulatory review times. Phase attrition and approval rates are the most important sources of variability in total clinical period costs between therapeutic categories (3). Therefore, the challenge faced by pharmaceutical industries is to minimize the hit-to-drug timeline improving costs. In this context, in silico computational therapeutics (or in silico pharmacology) are providing advances in medicine and therapeutics (2).
Molecules mutual recognition allows communication in biological systems. As a consequence, the geometrical and chemical complementarity of molecules (ligands) and their biological target structures (receptors) influences metabolic or signal transduction pathways initiating a physiological or therapeutic effect (4). Therefore, the identification of potent, high-affinity ligands is of utmost importance for the great expansion of validated targets expected from genomics for different diseases (5, 6). In contrast to technology-driven high-throughput screening (HTS) (HTS success depends on the assay system relevant to in vivo conditions and they are physically limited by the repertoire of compounds), rational drug design is a knowledge-driven approach of finding new medications that require structural information either on bioactive ligands for the target of interest (ligand-based virtual screening) or on the target itself (target-based virtual screening, often referred to as docking). Currently, the Protein Data Bank (PDB) holds more than 80,000 three-dimensional (3D) structures, but even this high number is insignificant. Some of the biomolecules have more than one bound to different molecules, and 3D structures of many important targets are still unknown. Similarly, the number of lead drug-like molecules is also relatively less (6). There are a very large number of known compounds and new ones are synthesized or discovered every year and the converging line of progress in chemistry and biology has generated a flood of information and knowledge that has gone beyond the usual capacity of “in cerebro” data handling, being a driving force of computer sciences (7). The process of drug discovery requires a multidisciplinary approach to really constitute the basis of a rational drug design. Here, we will review, in a simple way, some of the most popular strategies used by modelers and some successful applications on computational chemistry to raise awareness of its importance and potential.
Rational drug design
The strategy to be followed in rational drug design depends on whether the 3D structure of the biological target molecule is known or not (Figure 1). If the structure of the target is not known, a diverse range of ligand-based virtual screening methods exist, and they are all based on the central similarity-property principle which states that similar molecules should exhibit similar properties (structure-activity relationship or SAR) (8), and thus chemical similarity calculations are at the core of ligand-based virtual screening (9). The molecules in a particular database can be scored relative to similarity to one or multiple bioactive ligands and then ranked to reflect decreasing probability of being active, a procedure generally better than a random selection of molecules. Then, the top scoring molecules selected can be prioritized for going into experimental testing, representing a cost-effective strategy of improvement.
The birth of quantitative structure-activity relationships (QSARs) takes place when Hansch, in the 1950s, used calculators and statistics to arrive at quantitative relations between structure and a biological effect (10). The results of QSAR methods provide not only information about the structural requirement of the receptor responsible of that activity but also allow prediction of the activity of unknown compounds. An alternative and complementary procedure, and probably the most popular one, is the generation of a pharmacophore. By definition, a pharmacophore is “the ensemble of steric and electronic features that is necessary to ensure the optimal supramolecular interactions with a specific biological target structure and to trigger (or to block) its biological response. A pharmacophore does not represent a real molecule or a real association of functional groups, but a purely abstract concept that accounts for the common molecular interaction capacities of a group of compounds towards their target structure. The pharmacophore can be considered as the largest common denominator shared by a set of active molecules. This definition discards a misuse often found in the medicinal chemistry literature which consists of naming as pharmacophores simple chemical functionalities such as guanidines, sulfonamides or dihydroimidazoles (formerly imidazolines), or typical structural skeletons such as flavones, phenothiazines, prostaglandins or steroids” (11). In the absence of a known receptor structure, a pharmacophore can be identified from a set of ligands that have been observed to interact with the target receptor. The strength of pharmacophore-based screening compared with other ligand similarity screening approaches lies in the ability to detect a diverse set of putative active compounds with totally different chemical scaffolds. Successful applications of the use of pharmacophores in virtual screening include the identification of hits for a variety of targets such as protein kinase C (12) or HIV integrase (13, 14). In addition to screening, pharmacophore is also a powerful model in other applications of drug development, such as de novo design, lead optimization, ADMET (absorption-distribution-metabolism-excretion-toxicity) studies and chemogenomics (15–17).
A key step in computational 3D pharmacophore methods is the problem of pharmacophore elucidation: the identification from a set of active molecules and their biological activities of key common features and their relative orientations (also called pharmacophore mapping). Pharmacophore elucidation is a molecular alignment problem, the aim being to superimpose a set of active ligands, all of which bind to the same protein of unknown 3D structure, so that the features they have in common become evident (18). Programs for pharmacophore elucidation commercially available include PHASE (19), CATALYST (20), GALAHAD (21), GASP (22) and the pharmacophore module of MOE (23). In essence, all pharmacophore elucidation algorithms must include methods for representing the ligand, searching for candidate alignment and then scoring those alignments. This is not an easy task because a molecule must be considered or partitioned into a set of features (hydrogen-bond donor features, hydrogen-bond acceptor features, donor/acceptor features, atoms with formal charges or hydrophobic features), some of which can be more important than others depending on a particular case to build the pharmacophore. Moreover, searching for the candidate alignments molecular flexibility must be taken into account and, therefore, algorithms may operate on a pregenerated set of conformations for each ligand or alter ligand conformation on the fly during the alignment process. The scoring functions must also consider feature matching, volume overlap, strain energy and selectivity, and a weight at the end must be assigned to the different terms, which inevitably introduce arbitrariness and difficulty in obtaining a high-quality result (18).
When the structure of the target protein is known, receptor-based or target-based computational methods can be employed and different strategies are possible: in de novo design, novel leads are generated in the binding pocket starting from prepositioned seed atoms or fragments that are gradually grown into entire molecules (24, 25). Desirable compounds from the de novo design have to be synthesized normally. Alternatively a compound library can be screened for ligands in agreement with binding site requirements. The individual molecules are docked into the binding site (26). This docking strategy is also called virtual screening (27). Large virtual libraries of compounds are reduced in size to a manageable subset of possible candidate molecules, which, if successful, includes molecules with high binding affinities to a target receptor. Docking calculations have been applied in pharmaceutical research for almost three decades. The docking process (Figure 2) involves the prediction of ligand conformation and orientation (or posing) within a targeted binding site and the evaluation of the different poses or scoring. The accurate prediction of the binding mode of the selected molecule and its target (the docking problem) is of utmost importance and its solution lies in correctly computing the combination of enthalpic and entropic effects that come from the formation and destruction of interactions among the protein, ligand, and solvent in the form of hydrogen bonds, van der Waals interactions, formally charged interactions, and the entropy losses of the protein and ligand balanced against the entropy gains of the solvent. An accurate picture of a protein/ligand interaction would involve an ensemble of the most probable protein and ligand conformations given an accurate calculation of the free energies attributable to each state (28).
The protein-ligand docking process begins with the application of docking algorithms that pose the molecules in the active site. To accurately place the molecule in the active site, treatment of ligand flexibility can be done by using different search methods, which can be divided into three categories:
Systematic methods: systematic search algorithms are normally used for flexible ligand docking, which generate all possible ligand binding conformations by exploring all degrees of freedom of the ligand. There are three types of systematic search methods: exhaustive search, fragmentation and conformational ensemble. Exhaustive search algorithms are methods in which flexible-ligand docking is performed by systematically rotating all possible rotatable bonds of a ligand at a given interval. Glide (29, 30)FRED (31) are two practical examples of this type of methods. In fragmentation methods, the ligand is divided in rigid core fragments and in flexible parts. The core fragments are first docked into the binding site and the flexible parts of the ligand are added in an incremental way. This is a good approach when the rigid core has significant binding interactions. DOCK (32), HammerHead (33) and FlexX (34) are examples of docking programs that use fragmentation methods. In conformational ensemble methods, ligand flexibility is represented by rigidly docking an ensemble of pregenerated ligand conformations with other programs such as OMEGA (OpenEye Scientific Inc., Santa Fe, NM, USA). Then, ligand binding modes from different docking runs are collected and ranked according to their binding energies scores.
Random or stochastic methods: these algorithms make random changes in both conformational space and translational/rotational space to either a single ligand or a population of ligands. The three most used methods are Monte Carlo, e.g., MCDOCK, Tabu Search and Genetic algorithms. The key issue is to ensure that enough iterations are performed to sample the active site properly. The random change will be accepted or rejected according to probabilistic criterion.
Simulation methods: the most popular one would be molecular dynamics (MD), although energy minimization methods are gaining importance. MD programs such as AMBER, CHARMM and GROMOS provide the advantage of full simulation, where target and ligand are both treated as being flexible. However, MD simulations are often unable to step over high energy conformational barriers, leading to “trapping” and inadequate sampling. Nevertheless, an extension of MD simulations based on the replica-exchange framework (REMD) known as Hamiltonian REMD facilitates the characterization of conformational equilibria across large energetic barriers or in the presence of substantial entropic effects (4, 35–39).
Algorithms are complemented by scoring functions that are designed to predict the biological activity through the evaluation of interactions between compounds and potential targets. The scoring function is a key element of a protein-ligand docking algorithm, because it directly determines the accuracy of the algorithm. An ideal scoring function would be computationally efficient and reliable. Three classes of scoring functions are mainly applied:
Force field based scoring, which usually quantify the sum of energies, receptor-ligand interaction energy and internal ligand energy (most often described by using van der Waals and electrostatic energies terms), generally considering a single protein conformation which makes it possible to omit the calculation of internal protein energy simplifying the scoring, although these functions do not include solvation and entropic terms. One of the major challenges in force field scoring functions is how to account for the solvent. The simplest method is to use a distance-dependent dielectric constant such as the scoring function in DOCK:
where rij stands for the distance between protein atom i and ligand atom j, Aij and Bij are Van der Waals (VDW) parameters and qi and qj are atomic charges, ε(rij) is usually set to 4 rij, reflecting the screening effect of water on electrostatic interactions.
The most rigorous force field methods are to treat water molecules explicitly such as free-energy perturbation (FEP) and thermodynamic integration (TI). In addition to the challenge on solvent effect, how to accurately account for entropic effect is an even more severe challenge for force field scoring functions. These methods are normally computationally expensive.
Empirical scoring functions also called regression-based approaches are based on a set of weighted energy terms (VDW energy, electrostatic energy, hydrogen bonding energy, desolvation term, entropy term, hydrophobicity term, etc.) whose coefficients are derived by reproducing the binding affinity data of a training set of protein-ligand complexes with known 3D structures.
where ΔGi represents individual empirical energy terms, and the corresponding coefficients Wi. Although the empirical scoring function is computationally efficient because of its simple energy forms, its general applicability is training set-dependent.
Knowledge-based scoring functions that are designed to reproduce experimental structures rather than binding energies. In knowledge-based functions, protein-ligand complexes are modeled using relatively simple atomic interaction-pair potentials. The principle behind knowledge-based scoring functions is the potential of mean force which is defined by the inverse Boltzmann relation:
where kB is Boltzmann’s constant, T is absolute temperature of the system, ρ(r) is the number density of the protein-ligand atom pair at distance r in the training set, and ρ*(r) is the pair density in a reference state where the interatomic interactions are zero.
One major challenge for these scoring functions is the calculation for the previously mentioned state of reference. Traditional methods to approximate the reference state are randomization of the atoms in the training set, with the disadvantages of volume and interatomic connectivity exclusion among others. Recently, Haung and Zou have presented a new computational model (ITScore/SE) that explicitly includes contribution from solvation and entropy in the knowledge-based scoring functions (one of the most important limitations of these functions is that they do not explicitly include the contributions of solvation and entropy, parameters that play a critical role in determining the binding free energy between protein and ligand) based on a novel ITScore iterative method previously published. In the newly developed ITScore/SE, the solvation effect was included by using an atom-based solvent accessible surface area (SASA) term, and the entropic contribution was estimated by two empirical energy terms.
Consensus scoring, where it is possible to combine different scoring functions to balance errors to improve the probability of identifying “true” ligands (4, 36–39).
Applications of in silico pharmacology
The number of proteins with a known 3D structure is increasing rapidly and structures produced by structural genomics are being made available and openly accessible for use in drug discovery and development processes (40, 41). In silico methods have proved useful for successful rational drug design of amyloid aggregation inhibitors as potential Alzheimer’s disease drugs (42, 43)topoisomerase inhibitors (44), angiogenesis inhibitors (45) and a Csn-B-derived Nur77 agonist (46) as cancer therapeutics, 5-HT2C-selective ligands as schizophrenia drugs (47), and countless other promising examples (48–50)which are able to show how computational tools have become an integral aspect of optimizing drug design and development.
A successful virtual screening and docking exercise that can be cited is the discovery of disrupting agents of MIF (macrophage migration inhibitory factor, a cytokine involved in inflammatory diseases and cancer) to its receptor CD74. To discover small molecule inhibitors of the biological activity of MIF, virtual screening was performed by docking 2.1 million compounds from the ZINC (34) and Maybridge (www.maybridge.com) databases into the MIF tautomerase active site. After visual inspection of 1200 top-ranked MIF-ligand complexes, 26 possible inhibitors were selected and purchased and 23 of them were assayed. The in vitro binding assay for MIF with CD74 revealed that 11 of the compounds have inhibitory activity in the μM regime including four compounds with IC50 values below 5 μM. Inhibition of MIF tautomerase activity was also established for many of the compounds with IC50 values as low as 0.5 μM; Michaelis-Menten analysis was performed for two cases and confirmed competitive inhibition (51).
Other studies demonstrate that it is possible to learn from a formally unsuccessful virtual screening exercise and with the aid of computational analyses, to evolve a false-positive into a true active, virtual screening of the Maybridge library of ca. 70,000 compounds was performed to seek potential non-nucleoside inhibitors of HIV-1 reverse transcriptase. Four selected compounds were submitted after docking to an anti-HIV assay infected human T cells and found no ability to inhibit HIV replication. However, the highest ranked library compound was subjected to further computational analysis to seek modifications. A small substituent scan was performed with the BOMB program (Biochemical and Organic Model Builder, a ligand-growing de novo design program), which specifically suggested some changes on the core of the molecule. Synthesis and assaying of such compounds did yield active anti-HIV agents with EC50 values as low as 310 nM (52).
Computational methods are currently also being used to assess the ADME and ADMET properties of compounds at the early stages of discovery/development. Traditionally, drugs were discovered by testing compounds in a battery of in vivo biological screens, and then promising compounds were then further studied in development, where their pharmacokinetic properties, metabolism and potential toxicity were investigated. Adverse findings were often made at this stage, with the result that the project would be halted or restarted to find another clinical candidate (an unacceptable burden on the research cost of any pharmaceutical company) (53). Today, with computational methods, the testing of drug metabolism, pharmacokinetic profile of a compound and/or its toxicity is done much earlier. The need for early consideration of ADME properties is also very important due to the implementation of combinatorial chemistry and HTS, because this can generate vast numbers of potential lead compounds (54). In a recent review, Gleeson and colleagues analyze the different in silico models and QSAR rules available for the most popular ADME endpoints and their utilities in drug discovery (55). QikProp is among the earliest ADME programs that predicted a substantial array of pharmacologically relevant properties. Version 1.0 provided predictions for intrinsic aqueous solubility, Caco-2 cell permeability and several partition coefficients including octanol/water. Updated QikProp 3.0 version covered 18 quantities including log BB for brain-blood partitioning, log Khsa for serum albumin binding and primary metabolites. For example, QikProp 3.0 was used to process ca. 1700 known, neutral oral drugs. Consistent with the log Po/w limit of 5 in Lipinski’s rules, 91% of oral drugs are found to have QP log P values below 5.0. For aqueous solubility, 90% of the QP log S values are above –5.7, that is, S is <1 μM. The QikProp results also state that 90% of oral drugs have cell permeabilities, P Caco, above 22 nm/s and no more than six primary metabolites. These quantities address important components of bioavailability, namely, solubility, cell permeability and metabolism. In a lead optimization process, this ADME filter program would be the next step of the process after more rigorous FEP calculations to synthesize and assay the potential lead. Actually, a compound would be viewed as potentially problematic if it does not satisfy a “rule-of-three”: predicted log S>–6, P Caco>30 nm/s and maximum number of primary metabolites of 6. For activity requiring blood-brain barrier penetration, the predicted log BB should also be positive (56).
3D pharmacophores can also be derived from the binding site of a protein by using the direct observation of specific interactions between protein and ligand(s). Such pharmacophores can then be used in the usual way, to search a 3D database to identify compounds for focused screening or to create a much smaller subset database for docking. The distinction between “structure-based” and “ligand-based” methods is more diffuse and traditional ligand-based methods such as 3D pharmacophores can greatly enhance the efficiency and effectiveness of structure-based design, and many examples can be found in the literature, such as the identification of novel cannabinoid CB1 receptor antagonists (57), novel angiotensin converting enzyme 2 inhibitors (58) or lead optimization of non-steroidal glucocorticoid agonist (59).
There are a very large number of known compounds and new ones synthesized and discovered every year with massive amounts of physical and chemical property data generated each year according to their studies; combinatorial libraries have expanded the size of compound collections; HTS has enabled the screening of millions of compound libraries and functional genomics research has led to the identification of an unprecedented number of potential therapeutic protein targets indicating the need for an electronic information processing for storing, organizing, acquiring and evaluating data to give researchers a better overview of known chemistry (60, 61). The development of a new drug is an extremely complex, time-consuming process with an average cost of over $900 million; consequently, predictive methodologies for increasing productivity of research and development processes become more and more important for large pharmaceutical companies (1). Computational tools offer the advantage of delivering new drug candidates more quickly and at a lower cost. The major roles of computation in drug discovery are: (i) virtual screening and de novo design, (ii) in silico ADME/ADMET prediction and (iii) advanced methods for determining protein-ligand binding and structure-based drug design (62). We have given an overview of important aspects of rational drug design based on computational methods and how different approaches can be done taking into account if the target is not known (ligand-based methods) or known (target-based methods), as well as some successful applications. Two essential components of the docking process have been illustrated: docking or sampling (generation of putative ligand binding orientation/conformations near a binding site of a protein) and scoring (prediction of the binding tightness for individual orientations/conformations with a physical or empirical energy function). Currently, huge efforts are being given to both, especially to entropy and desolvation effects that remain the two major challenges for current scoring functions (38).
Nevertheless, because structural information and binding data about receptor-ligand complexes are constantly increasing and considering that most disease processes and disease treatments are manifest at the protein level, understanding the interactions between drugs and targets are of utmost importance to identify or optimize new hits or leads, or to redesign a molecule when performing in a comparative manner between diseased or healthy samples if target protein adopts different conformations (63).
The creation of a comprehensive, continuously updated open database for the human proteome-drug interactome (the complete map of protein interactions that can occur in a living organism is called the interactome) (64), including their predicted affinities and binding energies to each human protein, and/or open software to make accurate predictions, would be an indispensable tool to take the whole potential of computational methods, as a natural extension phase for the Human Genome Project (65). In silico pharmacology would never be able to entirely replace experimental approaches in vitro and in vivo but they have become part of an improved and effective drug discovery process.
The authors wish to thank Professor Romeo Romagnoli from University of Ferrara for his support and encouragement to young researchers’ initiatives.
Conflict of interest statement
Authors’ conflict of interest disclosure: The authors stated that there are no conflicts of interest regarding the publication of this article.
Research funding: None declared.
Employment or leadership: None declared.
Honorarium: None declared.
Verbanac D. Predictive methods as a powerful tool in drug discovery. Biochem Med 2010;20:314–8. [CrossRef]
Khan F, Yadav DK, Maurya A, Srivastava SK. Modern methods and web resources in drug design and discovery. Lett Drug Design Discov 2011;8:469–90. [CrossRef]
Dimasi JA, Hansen RW, Grabowski HA, Lasagna L. Research and development cost for new drugs by therapeutic category. A study of the United States pharmaceutical industry. Pharmacoeconomics 1995;7:152–69. [CrossRef] [PubMed]
Dean PM. Molecular recognition: the measurement and search for molecular similarity in ligand-receptor interaction. In: Maggiora GM, Johnson MA, editors. Concepts and applications of molecular similarity. New York: John Wiley & Sons, 1990:99–117.
Willett P. Similarity-based approaches to virtual screening. Biochem Soc Trans 2003;31:603–6. [PubMed]
Hansch C, Fujita T. Rho-Sigma-Pi analysis. Method for correlation of biological activity–chemical structure. J Am Chem Soc 1964;86:1616–26. [CrossRef]
Wermuth CG, Ganellin CR, Lindberg P, Mitscher LA. Glossary of terms used in medicinal chemistry. Pure Appl Chem 1998;70:1129–43. [CrossRef]
Wang SM, Zaharevitz DW, Sharma R, Marquez VE, Lewin NE, Du L, et al. The discovery of novel, structurally diverse protein kinase C agonist through computer 3D-database pharmacophore search. Molecular modeling studies. J Med Chem 1994;37:4479–89. [CrossRef]
Carlson HA, Masukawa KM, Rubins K, Bushman FD, Jorgensen WL, Lins RD, et al. Developing a dynamic pharmacophore model for HIV-1 integrase. J Med Chem 2000;43:2100–14. [PubMed]
Nicklaus MC, Neamati N, Hong HX, Mazumder A, Sunder S, Chen J, et al. HIV-1 integrase pharmacophore: discovery of inhibitors through three-dimensional database searching. J Med Chem 1997;40:920–9. [PubMed]
Ekins S, Swaan PW. Development of computational models for enzymes, transporters, channels and receptors relevant to ADME/Tox. Rev Comp Chem 2004;20:333–415.
Dixon SL, Smondyrev AM, Knoll EH, Rao SN, Shaw DE, Friesner RA. 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 2006;20:647–71. [CrossRef] [PubMed]
Richmond NJ, Abrams CA, Wolohan PR, Abrahamian E, Willet P, Clark RD. GALAHAD: 1. Pharmacophore identification by hypermolecular alignment of ligands in 3D. J Comput AidedMol Des 2006;20:567–87.
Chemical Computing Group. Molecular operating environment. Montreal, Canada: Chemical Computing Group (http://www.chemcomp.com/).
Murcko MA. Recent advances in ligand design methods. Rev Comput Chem 1997;11:1–66.
Clark DE, Murray CW, Li J. Current issues in de novo molecular design. Rev Comput Chem 1997;11:67–125.
Jaim AN. Effects of protein conformation in docking: improved pose prediction though protein pocket adaptation. J Comput Aided Mol Des 2009;23:355–74
Friesner RA, Banks JL, Murphy RB, Halgren TA, Klicic JJ, Mainz DT, et al. Glide: a new approach for rapid, accurate docking and scoring. 1. Method and assessment of docking accuracy. J Med Chem 2004;47:1739–49.
Ewing TJ, Kuntz ID. Critical evaluation of search algorithms for automated molecular docking and database screening. J Comput Chem 1997;18:1175–89. [CrossRef]
Faraldo-Gómez JD, Roux B. Characterization of conformational equilibria through Hamiltonian and temperature replica-exchange simulations: assessing entropic and environmental effects. J Comput Chem 2007;28:1634–47. [CrossRef]
Westbrook J, Feng ZK, Chen L, Yang HW, Berman HM. The protein data bank and structural genomics. Nucleic Acid Res 2003;31:489–91. [CrossRef]
Scherzer-Attali R, Pellarin R, Convertino M, Frydman-Marom A, Egoz-Matia N, Peled S, et al. Complete phenotypic recovery of an Alzheimer’s disease model by a quinone-tryptophan hybrid aggregation inhibitor. PloS One 2010;5:e11101. [CrossRef]
Takahashi T, Ohta K, Mihara H. Rational design of amyloid β peptide-binding proteins: pseudo-Aβ β sheet surface presented in green fluorescent protein binds tightly and preferentially to structured Aβ. Proteins 2010;78:336–47. [PubMed] [CrossRef]
Huang H, Chen Q, Ku X, Meng L, Lin L, Wang X, et al. A series of α-heterocyclic carboxaldehyde thiosemicarbazones inhibit topoisomerase II α catalytic activity. J Med Chem 2010;53: 3048–64. [CrossRef] [PubMed]
Murphy EA, Shields DJ, Stoletov K, Dneprovskaia E, McElroy M, Greenberg JI, et al. Disruption of angiogenesis and tumor growth with an orally active drug that stabilizes the inactive state of PDGFRβ/B-RAF. Proc Natl Acad Sci USA 2010;107: 4299–304. [CrossRef]
Liu JJ, Zeng NH, Zhang LR, Zhan YY, Chen Y, Wang YA, et al. A unique pharmacophore for activation of the nuclear orphan receptor Nur77 in vivo and in vitro. Cancer Res 2010;70: 3628–37. [CrossRef]
Kozikowski AP, Cho SJ, Jensen NH, Allen JA, Svennebring AM, Roth BL. HTS and rational drug design to generate a class of 5-HT2C-selective ligands for possible use in schizophrenia. Chem Med Chem 2010;5:1221–5. [CrossRef]
Coupez B, Lewis RA. Docking and scoring. Theoretically easy, practically impossible? Curr Med Chem 2006;13:2995–3003. [PubMed]
Geppert T, Bauer S, Hiss JA, Conrad E, Reutlinger M, Schneider P, et al. Immunosuppressive small molecule discovered by structure-based virtual screening for inhibitors of protein-protein interactions. Angew Chem Int Ed Engl 2012;51: 258–61. [CrossRef] [PubMed]
Cournia Z, Leng L, Gandavadi S, Du X, Bucala R, Jorgensen WL. Discovery of human macrophage migration inhibitory factor (MIF)-CD74 antagonists via virtual screening. J Med Chem 2009;52:416–24. [PubMed] [CrossRef]
van de Waterbeemd H, Gifford E. ADMET in silico modelling: towards prediction paradise? Nat Rev Drug Discov 2003;2: 192–204. [CrossRef]
Hodgson J. ADMET – turning chemical into drugs. Nat Biotechnol 2001;19:722–6. [CrossRef]
Gleeson P, Hersey A, Hannongbua S. In silico ADME models: a general assessment of their utility in drug discovery applications. Curr Topics Med Chem 2011;11:358–81. [CrossRef]
Wang HW, Duffy RA, Boykow GC, Chackalamannil S, Madison VS. Identification of novel cannabinoid CB1 receptor antagonists by using virtual screening with a pharmacophore model. J Med Chem 2008;51:2439–46.
Rella M, Rushworth CA, Guy JL, Turner AJ, Langer T, Jackson RM. Structure-based pharmacophore design and virtual screening for novel angiotensin converting enzyme 2 inhibitors. J Chem Inf Model 2006;46:708–16. [CrossRef] [PubMed]
Clackers M, Coe DM, Demaine DA, Hardy GW, Humphreys D, Inglis GG, et al. Non-steroidal glucocorticoid agonist. The discovery of aryl pyrazoles as A-ring mimetics. Bioorg Med Chem Lett 2007;17:4737–45. [CrossRef] [PubMed]
Taft CA, Da Silva VB, Da Silva CH. Current topics in computer-aided drug design. J Pharm Sci 2008;97:1089–98. [CrossRef]
Shukla S, Choubey SK, Srivastava P, Gomase VS. Chemoinformatics – an emerging field for computer aided drug design. J Biotechnol Lett 2010;1:10–4.
Rao VS, Srinivas K. Modern drug discovery process: an in silico approach. J Bioinform Seq Anal 2011;2:89–94.
Cusick ME, Klitgord N, Vidal M, Hill DE. Interactome: gateway into systems biology. Hum Mol Genet 2005;2: 171–81. [CrossRef]
Gurwitz D. Cataloging the interactome of small molecules and the human proteome. Drug Dev Res 2011;72:1–3. [CrossRef]