Content uploaded by Alexis Ivanov
Author content
All content in this area was uploaded by Alexis Ivanov on Sep 23, 2014
Content may be subject to copyright.
Current Drug Targets - Infectious Disorders, 2003, 3, 33-40 33
1568-0053/03 $41.00+.00 © 2003 Bentham Science Publishers Ltd.
Strategy of Computer-Aided Drug Design
A.V. Veselovsky* and A.S. Ivanov
V. N. Orekhovich Institute of Biomedical Chemistry, Russian Academy of Medical Science,
Pogodinskaya str., 10, Moscow, 119121, Russia
Abstract: Modern strategies of computer-aided drug design (CADD) are reviewed. The
task of CADD in the pipeline of drug discovery is accelerating of finding the new lead
compounds and their structure optimization for the following pharmacological tests. The
main directions in CADD are based on the availability of the experimentally determined
three-dimensional structure of the target macromolecule. If spatial structure is known the
methods of structure-based drug design are used. In the opposite case the indirect
methods of CADD based on the structures of known ligands (ligand-based drug design)
are used. The interrelationship between the main directions of CADD is reviewed. The
main CADD approaches of molecule de novo design and database mining are described. They include methods
of molecular docking, de novo design, design of pharmacophore and quantity structure-activity relationship
models. New ways and perspectives of CADD are discussed.
Key words: computer-aided drug design, structure-based drug design, ligand-based drug design, database mining, de novo
design, molecular docking, pharmacophore model, QSAR
1. INTRODUCTION
The pipeline of drug discovery from idea to market
consists of seven basic steps: disease selection, target
selection, lead compound identification, lead optimization,
preclinical trial testing, clinical trial testing and
pharmacogenomic optimization [1]. In practice, the last five
steps required to pass repeatedly. The compounds for testing
can be obtained from natural sources (plants, animals,
microorganisms) and by chemical synthesis. These
compounds can be rejected as perspectiveless owing to
absence or low activity, existence of toxicity or
carcinogenity, complexity of synthesis, insufficient
efficiency, etc. As a result only one of 100000 investigated
compounds may be introduced to the market, and now
average cost of development of new drug rose up to 800
million dollars [2, 3]. The reduction of time-consuming and
cost of the last stages of drug testing is unlikely due to strict
state standards on their realization. Therefore main efforts to
increasing efficiency of development of drugs are directed to
stages of discovery and optimization of ligands.
Extensive genome decoding of various organisms,
including man, proteomic investigations, discoveries of
molecular mechanisms of many diseases, advances of protein
chemistry lead to dramatic increase of number of new
potential targets. The present stage of pharmacology is
characterized as “target-rich lead-poor environment” [1]. So
we may expect, that in the nearest future discovery of new
ligands for new targets will be the weakest point in
pharmacology.
Address correspondence to this author at the V. N. Orekhovich Institute of
Biomedical Chemistry, Russian Academy of Medical Science,
Pogodinskaya str., 10, Moscow, 119121, Russia; Fax: 095-245-0857; Email:
veselov@ibmh.msk.su
During the last decades the field of drug discovery
process that direct to new ligands finding turns into the
modern science employing of computer, bioinformatic and
experimental approaches, which are denominated as «rational
drugs design». The latter includes two directions:
experimental and computer methods (CADD - computer
aided drug design) [4-6]. The main experimental methods are
combinatorial chemistry [7] and high-throughput screening
[8].
Computer methods of drugs design are based on a
postulate that pharmacologically active compounds act by
interaction with their macromolecule-targets, mainly proteins
or nucleic acids. Major factors of such interaction include
steric complementarity of interacting surfaces of molecules,
electrostatic force, hydrophobic interaction and hydrogen
bonds formation. These factors are mainly considered during
analysis and prediction of interaction of two molecules.
2. HARDWARE AND SOFTWARE REQUIREMENT
Modeling of interaction of macromolecules with ligands
requires various methods of calculations. Such studies
usually employ multiprocessor computer systems under
UNIX management [9]. At present commercial software
packages SYBYL (Tripos Inc.) (http://www.tripos.com) [10]
and Insight II (Accelrys) (http://www.accelrys/insight) are
widely used for these purposes. These software packages
allow carrying out the almost complete process of drug
design. For more limited purposes other commercial
programs can also be used: HyperChem
(http://www.hyper.com/products/), programs of ACD
(http://www.acdlabs.com/products), etc. Also there are
numerous freeware and shareware molecular modeling
34 Current Drug Targets - Infectious Disorders, 2003, Vol. 3, No. 1 Veselovsky and Ivanov
Fig. (1). Pathways of computer aided drug design.
programs directed for local purposes. Also it is crucial to
have access to the databases with structures of
macromolecules and small compounds. The spatial
structures of proteins are assembled in Protein Data Bank
(PDB) (http://www.rcsb.org/pdb) [11], and small
compounds in Cambridge Structural Database (CSD)
(http://www.ccdc.cam.ac.uk/prods/csd/csd.html). There are
many other databases of low-weight compounds, but
databases with commercially available compounds are the
most convenient for practice, because these compounds can
be quickly acquired for experimental testing.
3. SELECTION OF STRATEGY: TWO MAIN
QUESTIONS
At the beginning it is necessary to answer two basic
questions, which define the direction of further investigation
(Fig. (1)):
1) Is three-dimensional (3D) structure of the target is
available?;
2) What type of compounds is necessary?
Strategy of Computer-Aided Drug Design Current Drug Targets - Infectious Disorders, 2003, Vol. 3, No. 1 35
If spatial structure of target is known, the methods of
structure-based drug design (SBDD) (or direct methods) are
applicable. In this case compounds with complementary
properties to target surface may be designed, based on the
knowledge of properties and features of the spatial structure
of macromolecule. If such structure remains unknown,
methods of ligand-based drug design (LBDD) (or indirect
methods) are should be applied. In this case analysis of set
of ligands is carried out to reveal the common basic
properties of these molecules, which correlate with these
biological activity. Direct methods of drug design are more
preferable, since functional groups (which can interact with
ligands) and their spatial position are known. When the
spatial structure of the target is unknown, it is possible to
check the existence of known three-dimensional structures
among its relative protein-homologues. If 3D structure of
related homologues is available, it is possible to model the
target protein [6, 12]. However in this case model reliability
requires more than 35% identity between primary sequences
of target and homologue proteins. Now reliability of model
is determined by the main chain of protein, whereas the
structure of its active site is mainly determined by the
position of side groups of amino acids. Consequently, it is
necessary to employ template with high identity in the area
of the active site. This would allow to obtain rather reliable
model of the target for subsequent SBDD. Several attempts
to use such models for searching of new ligands have been
reported [13, 14].
The answer on the second question determines the
direction of investigation: search for new lead compounds or
optimization of known base structure for increase of activity
or decrease of side-effects.
4. STRUCTURE-BASED DRUG DESIGN
The first step of SBDD consists in analysis of 3D
structure of target and definition of ligand binding site. If
the spatial structure of complex of the target with known
ligand (substrate or competitive inhibitor) is available, the
binding site is determined unequivocally. Otherwise it is
necessary to detect it. This can be done by several ways. The
roughest approach is definition of cavities in protein
structure, and the biggest cavity seems to represents the
active site. More correct approach is searching the key amino
acids involved into catalysis (for example, triad of serine
protease) or cofactor, and region near them is predicted as the
binding site. However, problem of uncertainties of spatial
position of ligand at the active site still remains. Docking of
known substrates and/or inhibitors to enzyme represents the
optimal approach and on the basis of position of docked
ligands predicts the binding site of enzyme (see below about
docking).
There are two basic strategies for searching of biological
active compounds by SBDD: molecular database mining for
new ligands and ligand design de novo.
The modern preferential strategy for searching new lead
structures implies initial molecular database screening. The
cause is possibility for rather quick selection and testing of
numerous compounds. If compounds with required activity
have been discovered, they can be used for further
modifications for increasing activity, decreasing side-effects
and optimization of pharmacokinetic properties. Several
million of compounds (natural or synthetic) are known and
accumulated in various databases now. The method of
searching in molecular databases is based on the assumption
that compounds with requested activity are synthesized
earlier, but they have been not tested for this particular
activity. So they can be found in databases.
Several methods for molecular databases screening are
applicable [15]. The main of them is docking of compounds
in active site of target. Several docking programs have been
developed. DOCK is one of the widely used programs [16,
17]. In our laboratory the program DockSearch for fast
molecular geometrical docking was developed
(http://lmgdd.ibmh.msk.su/originalsoftware/DS/DS.htm). It
allows to screen large databases containing small
compounds. Among the programs for flexible docking of
small compounds AutoDock [18] and FlexX [19] should be
mentioned. The most algorithms of the docking programs
use rigid structure of molecule-target.
All docking programs generate hypotheses about
probable spatial positions of ligands in active site of the
target macromolecule. An estimation of results is carried out
by scoring functions (binding energy, area of contacting
surfaces, etc.). On the basis of these valuations the
compounds with best complementarity to structure and
properties of protein surface are selected [15, 20].
Usually preprocessing of molecular database is carried
out before the docking procedure. The compounds exhibiting
certain properties (molecular mass, presence of the certain
groups, similarity with known ligands, drug-likeness, etc.)
are selected.
The obvious advantage of this method consists in high
rate of obtaining results, ability to examine the numerous
variants. However, this approach is unable to discover the
compounds with high activity.
Several programs of ligand de novo design were
developed for designing new ligands and structure
optimization of known molecules. These include LUDI [21],
CLIX [22], CAVEAT [23], LeapFrog [10], etc. All of them
share similar principles, which consist in virtual modeling
of molecules, optimization of their structures and spatial
position in the active site with account of steric, electrostatic
interactions, hydrogen bonds formation and other factors
participating in protein-ligand interaction. The ligand design
by these programs begins from searching of specific groups,
capable to participate in noncovalent interactions with the
target macromolecule. The following step is selection of
linkers for fusion of these groups. Simultaneously
conformation of designed ligands and their spatial positions
are optimized. Binding energy is estimated after each step of
design, and the compounds with the highest values of
energy are selected for further optimization.
Efficiency of ligands binding to the target is estimated
by values of virtual binding energy. This is the major
problem for all methods of SBDD through our inability to
36 Current Drug Targets - Infectious Disorders, 2003, Vol. 3, No. 1 Veselovsky and Ivanov
Fig. (2). Pharmacophore model of HIV protease, which was designed on the basis of known 3D structures of HIV protease-inhibitor
complexes. A - Cross-section of the active site of HIV protease with superimposed inhibitors of HIV protease. B - The pharmacophore
model of HIV protease. This consists of three pharmacophore points (shown as balls): aromatic centroid (AC), acceptor atom (oxygen)
(AA) and proposed donor site on protein (DS). The inhibitor of HIV protease is shown as reference structure.
predict the real binding energy and converse these values
into experimentally determined values (IC50, Ki, Kd).
Although for several sets of compounds reasonable
correlation between structure and interaction energy were
found, the relationships are not generally transferable from
one protein system to the other [24]. The reason of this is
the limited amount of calculated dynamic parameters,
influence of the solvent and entropy effects, etc [25].
SBDD is finished by selection of molecules by predicted
energy interaction values and subsequent validation of their
activity in direct experiments.
Numerous inhibitors of HIV protease are a good example
of successful application of SBDD methods [26]. Other suc-
cessful examples are reviewed by Schneider and Bohm [27].
5. LIGAND-BASED DRUG DESIGN (LBDD)
The methods of this direction (Fig. (1)) are applied when
the spatial structure of macromolecule-target is unknown,
and there is unable to design its reliable model (see above).
These methods are based on analysis of sets of ligands with
known biological activity. They include: design of
pharmacophore models [28, 29], analysis of quantitative
structure-activity relationship (“classic” QSAR) [30] and its
modification 3D-QSAR (which takes into account spatial
structure of compounds) [31], quantitative structure-
properties relationship (QSPR) [32, 33], etc. Methods of
LBDD can be used for lead compounds discovery and
optimization of previously known ligands.
As for SBDD lead compounds discovery prefer to begin
by molecular databases mining. Pharmacophore models or
various models of cavity of the active site are used for this
purpose.
Pharmacophore model represents a set of points in space
with the certain properties and distances between them,
which define binding of given group of ligands with target.
Pharmacophore points can be the positively and negatively
charged atoms, cyclic groups, donor or acceptor atoms. Such
model designs by alignment of set of known ligands to
Strategy of Computer-Aided Drug Design Current Drug Targets - Infectious Disorders, 2003, Vol. 3, No. 1 37
Fig. (3). Models of moulds of substrate/inhibitor binding region of monoamine oxidase A (A) and monoamine oxidase B (B). These
moulds were designed by spatial alignment of sets of known effective competitive inhibitors of enzymes.
reveal best cluster of similar pharmacophore points of
different molecules in one space area. DISCO program from
SYBYL is an example of such type program (Fig. (2)).
Molecular database mining based on pharmacophore
models allows to select molecules that satisfy such model.
However, it is nearly impossible to find out compounds
from new class of chemical compounds; usually the selected
compounds represent analogues of known structures.
Other method for molecular database mining uses
docking of compounds into the model of spatial structure of
the active site of enzyme. These models are based on
analysis of structures and properties of known ligands.
Active analogues approach [34] and pseudoreceptor model
[35, 36] are examples of employment of such approaches.
We developed related approach of modeling of mould of
enzyme active site structure [37, 38]. This method is based
on using only effective ligands, which are well
accommodated within the active site. It results in the
description of cavity, in which active ligands can
accommodate. Figure (3) shows moulds of similar enzymes
(MAO A and MAO B) that are characterized by distinct
substrate and inhibitor specificity. It can be seen, that
designed moulds considerably differ from each other.
Docking compounds from molecular database into MAO A
mould with subsequent prediction of activity by 3D-QSAR
models (about 3D-QSAR models see bellow) allowed us to
select five compounds for testing and four of them have
appeared to be selective ÌÀÎ A inhibitors (prepared for
publication).
Prediction of activity of new compounds found in
databases and optimization of known structures are carried
out by methods of structure-activity relationship. The base
of “classical” QSAR method is the regression analysis of
relationship between biological activity of set of
homologues compounds and their various descriptors. The
correlation equations of this relationship allow to predict
activity for new analogues [30]. In these analysis descriptors
reflect various structural features (steric factors), electrostatic,
hydrophobic, donor-acceptor and other properties of
molecules.
Regression analysis for revealing relationship between
calculated descriptors and biological activity is carried out.
The reliability of model is estimated by statistical
parameters and correctness of prediction of test compounds
activity [30]. The method of neural networks has been also
applied for design models [39].
At present the method three-dimensional QSAR is widely
used. The special methods for the description of distribution
of ligand properties in three-dimensional space (CoMFA -
Comparative M olecular F ields A nalysis; CoMSIA -
Comparative Molecular Similarity Indices Analysis) are
applied [31, 40, 41]. This approach allows characterizing
steric, electrostatic, hydrophobic regions around molecules,
which define interaction of set of ligands with target. The
values of energy of interaction between ligand and test atom
for description of properties in space around molecules are
determined [31].
Then 3D-QSAR analyses are carried out by Partial Least
Square (PLS) analysis of the data of calculated values of
interaction energy [31]. The analysis of obtained 3D-QSAR
models is carried out by the contour map of different fields
showing favorable and unfavorable regions for interaction
around ligands.
38 Current Drug Targets - Infectious Disorders, 2003, Vol. 3, No. 1 Veselovsky and Ivanov
The indirect methods of modeling allow to estimate
probable pharmacological activity of unknown compounds.
The “classical” QSAR is effective for development of close
analogues of known compounds [42]. The 3D-QSAR
methods are somewhat capable to predict adequately
pharmacological activity for compounds from different
chemical classes [31]. The prediction of activity of
compounds by QSAR method is considered to be
satisfactory accuracy when the prediction is in the same order
of magnitude as real biological activity.
6. LINK WITH EXPERIMENTAL TESTING
(RUNNING ON THE CIRCLE)
The computer part of search and optimization of structure
of compounds is finished by passing these structures with
predicted activity to chemists for synthesis (if it is
necessary) or biochemists for testing for activity (if particular
compounds were found in available database) (Fig. (1)). The
experimental testing should answer the questions:
1) Whether the selected compounds show required
activity?;
2) And if they show it, is it enough for further
pharmacological testing?
If the selected compounds are inactive the second cycle of
computer modeling will be carried out taking into account
the obtained negative results (taking another molecular
database for mining, remodeling of pharmacophore or QSAR
models, additional analysis of structure of active site:
checking ability of conformational changing during ligand
interaction, participating molecules of water in ligand
binding, etc.). If testing compounds exhibit high (request)
activity, they can be subjected for the further tests (for
activity in vivo, toxicity, carcinogenity, etc.). If the
compounds demonstrate low activity in these tests, cycle of
optimization of structure for increase of their activity should
be carried out. For this purpose the programs of de novo
design (when the structure of the target is known) or QSAR
model (in the opposite case) are used. At this step the set of
related compounds with selected lead structure should be
synthesized and tested for biological (pharmacological)
activity. This cycle of optimization can be repeated several
times. Further the cycle of computer structure optimization
for pharmacokinetic properties optimization (adsorption,
distribution, metabolism, and excretion - ADME) [43] is
also possible.
7. NEW WAYS IN CADD
During last years 3D-QSAR models are designed for the
subsequent prediction of activity of compounds found in
molecular databases or designed de novo even when the
spatial structure of target is known. The reason for this
approach consists in significant difficulties of translation of
calculated virtual energies of interaction to real values (Kd,
IC50) (see above). So 3D-QSAR models are designed on
sets of known ligands using alignment rules chosen by
researchers [44, 45], or using alignment obtained by
preliminary docking of these ligands in the enzyme active
site [45, 46].
Initially computer methods of new ligands design and
discovery were developed for selection of ligands that
interact noncovalently with their targets. Recently new
approach for computer design of irreversible inhibitors has
been introduced. For this purpose it is proposed to use
compounds consisting of two parts: one part is reversibly
interact with target and the second part represents of middle
reactive group capable to covalently interact with certain
amino acid residues. The interaction of such molecule passes
in two steps. Preliminary the first part of molecule binds in
active site and oriented the reactive group closed to target
amino acid residue. As a result the covalent complex should
be formed with increasing inhibitory efficiency [47]. Such
compounds can be very useful for antimicrobial and antiviral
drugs.
Another way in computer drugs design is the
development of ligands for blocking protein-protein
interactions. Numerous enzymes act as a complex of several
subunits and all main cell functions occur through protein-
protein interaction. Therefore inhibition of enzyme assembly
from subunits or disruption of protein-protein interaction of
different proteins may result in change in action of these cell
systems. The contact surfaces of interacting proteins have
unique properties and can be considered as perspective
targets of new generation of drug [48-50]. The contacting
regions of proteins are usually rather conservative, since
single mutation in one contacted area can lead to disruption
of protein-protein interaction and accordingly cell system
action. So conservation of interaction requires
complementary mutation in another subunit that is rather
infrequent process. Hence, the occurrence of resistance for
inhibitors of protein-protein interaction is improbable
processes induced by spontaneous mutagenesis. The first
investigation directed on design of such type of inhibitors
appeared recently [48-51]. It can be peptides, their
modifications and small compounds. The design of small
specific inhibitors of protein-protein interactions can be done
by CADD methods.
8. CONCLUSION
The computer-aided drug design represents a complex
discipline using achievement in various areas of science and
various methods and approaches. It is directed to acceleration
and optimization of discovery of new biologically active
compounds. However, these approaches cannot replace the
experimental tests. The purpose of CADD is generation of
hypotheses about probable new ligands and their interaction
with targets. It is considered, that these methods can reduce
amount of the compounds that are needed to be synthesized
and tested for biological activity up to two orders. Thus,
they are capable to decrease essentially time-consuming and
financial expenses for development of drugs. Striking
example is given by J.Augen: “using traditional drug
development techniques it took nearly 40 years to capitalize
on a basic understanding of the cholesterol biosynthesis
pathway to develop statin drugs… Conversely, a molecular-
level understanding of the role of the HER-2 receptor in
Strategy of Computer-Aided Drug Design Current Drug Targets - Infectious Disorders, 2003, Vol. 3, No. 1 39
breast cancer led to the development of the chemotherapeutic
agent Herceptin® within only three years. The developers of
Herceptin® enjoyed the advantages of in silico molecular
modeling” [1]. It is necessary to note, that CADD is an
additional part of the first step of cycle of new drugs design.
In this sense, it is possible to assimilate this discipline as
enzymes (catalysts) in biochemical (chemical) reactions,
which can accelerate rate of reaction, reduce activation
barrier, but are not capable to change way of reaction.
It is necessary to note, that CADD needs several basic
data: three-dimensional structure of target, or its
homologues, and also sets of ligands. If such information is
not available the only way is experimental methods. The
high-throughput screening [8] is optimal method for
discovery of active compounds for new targets. However,
when the three-dimensional structure of target is known,
structure-based docking is preferred over random screening
[52].
The computer technologies can be also used in other
steps of drug development. The approaches for new
perspective targets for drugs based on bioinformatic analysis
of genome sequences of different organisms have appeared
[53-55]. There are numerous computer models for prediction
of various pharmacokinetic properties of compounds
(ADME) by their structure [43]. The expert systems, which
are capable to predict spectra of probable biological activity
of low-weight compounds, have been developed. Such
systems can be used as for selection of compounds for
experimental testing for requested activity, and for prediction
of possible side-effects (toxicity, cancerogenity, etc.). These
systems allow to find out unwanted negative properties of
the developing new drugs at early stages of preclinical trials.
PASS (http://www.ibmh.msk.su/PASS/) is an example of
such program [56].
At the first sight the methods and approaches employed
in CADD look perfect, simple and exhaustive, however, in
reality this direction is just developing. There are evident
successes in design of new biologically active substances,
some of which are applied in medical practice. However
significant efforts in development of new and modification
of existing tools are required for transformation of this area
from art into the advanced technology.
ACKNOWLEDGEMENTS
This work was partially supported by Russian
Foundation for Basic Research (grant 01-04-48128), INTAS
(grant 99-00433) and Regional Public Foundation for
Assistance to Domestic Medicine (Russia). The authors
thank Dr. A.E. Medvedev and Dr. V.S. Skvortsov for
valuable and helpful discussions.
ABBREVIATIONS
3D = Three-dimensional
ADME=Adsorption, distribution, metabolism, and
excretion
CADD = Computer-aided drug design
CoMFA = Comparative molecular fields analysis
CoMSIA = Comparative molecular similarity indices
analysis
CSD = Cambridge structural database
LBDD = Ligand-based drug design
PDB = Protein Data Bank
PLS = Partial Least Square
QSAR = Quantitative structure-activity relationship
SBDD = Structure-based drug design
REFERENCES
[1] Augen, J. Drug Discov. Today, 2002, 7, 315.
[2] Lohse, M.J. Trends Pharmacol. Sci., 1998, 19, 198.
[3] Borchardt, J.K. The Alchemist 6 December 2001,
http://www.chemweb.com/alchem/articles/10059288538
06.html.
[4] Marrone, T.J.; Briggs, J.M.; McCammon, J.A. Annu. Rev.
Pharmacol. Toxicol., 1997, 37, 71.
[5] Zeng, J. Comb. Chem. High Throughput Screen., 2000, 3,
355.
[6] Ooms, F. Curr. Med. Chem., 2000, 7, 141.
[7] Myers, P.L. Curr. Opin. Biotechnol., 1997, 8, 701.
[8] Fernandes, P.B. Curr. Opin. Chem. Biol., 1998, 2, 597.
[9] Guida, W.C. Curr. Opin. Struct. Biol., 1994, 4, 777.
[10] SYBYL® 6.7.1, Tripos Inc., 1699 South Hanley Rd., St.
Louis, Missouri, 63144, USA.
[11] Berman, H.M.; Westbrook, J.; Feng, Z.; Gilliland, G.;
Bhat, T.N.; Weissig, H.; Shindyalov, I.N.; Bourne, P.E.
Nucl. Acids Res., 2000, 28, 235.
[12] Amzel, L.M. Curr. Opin. Biotechnol., 1998, 9, 366.
[13] Belkina, N.V.; Skvortsov, V.S.; Ivanov, A.S.; Archakov,
A.I. Voprosy Meditsinskoi Khimii, 1998, 44, 464. (in
Russian).
[14] Yamamoto, K.; Masuno, H.; Choi, M.; Nakashima, K.;
Taga, T.; Ooizumi, H.; Umesono, K.; Sicinska, W.;
VanHooke, J.; DeLuca, H.F.; Yamada, S. Proc. Natl. Acad.
Sci. USA, 2000. 97, 1467.
[15] Weber, H.P. In Computer-Aided Drug Design Industrial
Research, Herrmann, E.C.; Franke, R., Eds. Springer
Verlag: Berlin. 1995, pp. 111-128.
[16] Kuntz, I.D.; Blaney, J.M.; Oatley, S.J.; Landridge, R.;
Ferrin, T.E. J. Mol. Biol., 1982, 161, 269.
40 Current Drug Targets - Infectious Disorders, 2003, Vol. 3, No. 1 Veselovsky and Ivanov
[17] Ewing T.J.A.; Makino S.; Skillman A.G.; Kuntz, I.D. J.
Comp.-Aided Mol. Design, 2001, 15, 411.
[18] Goodsell, D.S.; Morris, G.M.; Olson, A.J. J. Mol.
Recognit., 1996, 9, 1.
[19] Rarey, M.; Wefing, S.; Lengauer, T. J. Comp.-Aided Mol.
Design, 1996, 10, 41.
[20] Hoffmann, D.; Kramer, B.; Washio, T.; Steinmetzer, T.;
Rarey, M.; Lengauer, T. J. Med. Chem., 1999, 42, 4422.
[21] Bohm, H.J. J. Comp.-Aided. Mol. Design, 1992, 6, 61.
[22] Lawrence, M.C.; David, P.C. Proteins: Struct. Funct.
Genet., 1992, 12, 31.
[23] Bartlett, P.A.; Shea, G.T.; Telfer, S.J.; Waterman, S. In
Molecular recognition in chemical and biological
problems; Special Pub., Royal Chem.Soc.: London,
1989, Vol.78, pp. 182-196.
[24] Hubbard, R.E. Curr. Opin. Biotech., 1997, 8, 696.
[25] Norinder, U. In Computer-Aided Drug Design Industrial
Research, Herrmann, E.C.; Franke, R., Eds. Springer
Verlag: Berlin. 1995, pp. 99-109.
[26] Wlodawer, A.; Vondrasek, J. Annu. Rev. Biophys. Biomol.
Struct., 1998, 27, 249.
[27] Schneider, G.; Bohm, H.-J. Drug Discov. Today, 2002, 7,
64.
[28] Flohr, S.; Kurz, M.; Kostenis, E.; Brkovich, A.; Fournier,
A.; Klabunde. T. J. Med. Chem., 2002, 45, 1799.
[29] Ghose, A.K.; Wendoloski, J.J. In 3D QSAR in Drug
Design; Kubinyi, H. Ed.; Kluwer Academic Publishers:
Great Britain, 1998, Vol. 3, pp. 253-271.
[30] Kubinyi, H. Quant. Struct.-Act. Relat., 1994, 13, 285.
[31] Kim, K.H. In Molecular simulation and Drug Design;
Dean, P.M. Ed.; Blackie academic & professional:
London, 1995, pp. 291-331.
[32] Baumann, K. Trends Anal. Chem., 1999, 18, 36.
[33] Ponec, R.; Amat, L.; Carbo-Dorca, R. J. Comp.-Aided Mol.
Design, 1999, 13, 259.
[34] Kettmann, V.; Holtje, H.-D. Quant. Struct.-Act. Relat.,
1998, 17, 91.
[35] Zbinden, P.; Dobler, M.; Folkers, G.; Vedani, A. Quant.
Struct.-Act. Relat., 1998, 17, 122.
[36] Schleifer, K.-J. J. Comp.-Aided Mol. Design, 2000, 14,
467.
[37] Veselovsky, A.V.; Medvedev A.E.; Tikhonova, O.V.;
Skvortsov, V.S.; Ivanov, A.S. Biochemistry (Moscow),
2000, 65, 910.
[38] Veselovsky, A.V.; Tikhonova, O.V.; Skvortsov, V.S.;
Medvedev A.E.; Ivanov, A.S. QSAR SAR Environm. Res.,
2001, 12, 345.
[39] Manallack, D.T.; Livingstone, D.J. Eur. J. Med. Chem.,
1999, 34, 195.
[40] Cramer III, R.D.; Petterson, D.E.; Bunce, J.D. J. Am. Chem.
Soc., 1988, 110, 5959.
[41] Klebe, G. In 3D QSAR in Drug Design; Kubinyi, H. Ed.;
Kluwer Academic Publishers: Great Britain, 1998, Vol. 3,
pp. 87-104.
[42] Fujita, T. Quant. Struct.-Act. Relat., 1997, 16, 107.
[43] Clark, D.E.; Pickett, S.D. Drug Discov. Today, 2000, 5,
49.
[44] Hilgeroth, A.; Fleischer, R.; Wiese, M.; Heinemann, F.W.
J. Comp.-Aided Mol. Design, 1999, 13, 233.
[45] Sippl, W. J. Comp.-Aided Mol. Design, 2000, 14, 559.
[46] Sippl, W.; Contreras, J.-M.; Parrot, I.; Rival, Y.M.;
Wermuth, C.G. J. Comp.-Aided Mol. Design, 2001, 15,
395.
[47] Way, J.C. Curr. Opin. Chem. Biol., 2000, 4, 40.
[48] Cochran, A.G. Curr. Opin. Chem. Biol., 2001, 5, 654.
[49] Cody, W.L.; Lin, Z.W.; Panek, R.L.; Rose, D.W.; Rubin,
J.R. Curr. Pharm. Design, 2000. 6, 59.
[50] Veselovsky, A.V.; Ivanov, Yu.D.; Ivanov, A.S.; Archakov,
A.I.; Lewi, P.; Janssen, P. J. Mol. Recogn., 2002. 15, 1.
[51] Zutshi, R.; Brickner, M.; Chmielewski, J. Curr. Opin.
Chem. Biol., 1998, 2, 62.
[52] Doman, T.N.; McGovern, S.L.; Witherbee, B.J.; Kasten,
T.P.; Kurumbail R.; Stallings, W.C.; Connolly, D.T.;
Shoichet B.K. J. Med. Chem. 2002, 45, 2213.
[53] Dubanov, A.V.; Ivanov, A.S.; Archakov, A.I. Voprosy
Meditsinskoi Khimii, 2001, 47, 353. (in Russian).
[54] Freiberg, C. Drug Discov. Today, 2001, 6, S72.
[55] Spaltmann, F.; Blunck, M.; Ziegelbauer, K. Drug Discov.
Today, 1999, 4, 17.
[56] Poroikov, V.V.; Filimonov, D.A.; Borodina, Yu.V.;
Lagunin, A.A.; Kos, A. J. Chem. Inf. Comput. Sci. 2000,
40, 1349.
