This is the html version of the file
G o o g l e automatically generates html versions of documents as we crawl the web.

Google is neither affiliated with the authors of this page nor responsible for its content.
These search terms have been highlighted: piecewise linear potential dock 

Page 1
Lead optimization via high-throughput molecular docking
Diane Joseph-McCarthy
*, J Christian Baber
, Eric Feyfant
, David C Thompson
& Christine Humblet
Wyeth Research
Chemical and Screening Sciences
200 CambridgePark Drive
MA 02140
Wyeth Research
CN 8000
NJ 08543
*To whom correspondence should be addressed
Current Opinion in Drug Discovery & Development 2007 10(3):264-274
© The Thomson Corporation ISSN 1367-6733
Structure-based lead optimization approaches are increasingly
playing a role in the drug-discovery process. Recent advances
in 'high-throughput' molecular docking methods and examples
of their successful use in lead optimization are reviewed.
Measures of docking accuracy, scoring function comparisons,
and consensus approaches are discussed. Differences in docking
protocols typically used for lead optimization versus lead
generation are highlighted; this section includes a discussion of
the latest methods for the incorporation of protein flexibility.
New approaches developed specifically for the design of
combinatorial libraries as well as those designed or used for
'fragment' versus lead optimization are presented. Finally,
potential future improvements to the technology are outlined.
Keywords Combinatorial library design, computer-aided
drug design, fragment docking, scoring functions, structure-
based design, virtual screening
ADME absorption, distribution, metabolism and excretion,
ANOVA analysis of variance, ASP Astex statistical potential,
FEP free energy perturbation, GB generalized Born, HTS
high-throughput screening, IFD induced fit docking, LIE
linear interaction energy, MM molecular mechanics, MW
molecular weight, NMR nuclear magnetic resonance, PB
Poisson-Boltzmann, PDE phosphodiesterase, PLP piecewise
linear potential, PMF potentials of mean force, PTP1B
phosphotyrosine phosphatase 1B, RMSD root-mean-squared
deviation, ROC receiver operating characteristic, SA surface
area, SAR structure-activity relationships, vdw van der Waals
Structure-based lead optimization
Structure-based lead optimization is a powerful approach
employed by all large pharmaceutical and biotechnology
companies whenever structural information is available for
a project. The effectiveness of the approach and the exact
methods applied are determined by a number of factors: the
source of the initial lead, the quality of the lead, and the
accuracy and extent of the structural information available.
Computational chemistry techniques are utilized to
transform existing structural information into data that are
accessible and can be used to design improved compounds
or other leads by exploratory and medicinal chemistry
(Figure 1).
An initial lead can be generated via a variety of approaches:
by high-throughput screening (HTS), by more focused
screening of known inhibitors for a given target family, by
virtual screening of large databases of lead-like small
molecules, by modification of compounds that have been
previously published in the literature, by experimental
fragment screening, or by virtual fragment screening. The
quality of a lead in terms of its physical properties, potency
and target selectivity is often related to its source [1]. For
example, a validated HTS hit from a corporate collection
may be deemed too similar to related internal program lead
compounds, and any further optimization may require
scaffold modification to improve target selectivity against
other known therapeutic targets. Such modifications can
result in a reduction in potency. Virtual screening, in
particular, often generates weak hits that require further
optimization to improve potency. If vendor databases such
as ChemNavigator ( are used for
screening, novel functionality often has to be engineered
into the lead compound because the screened compounds
are in the public domain and available to many different
companies working on similar projects. Fragment screening
is often achieved via nuclear magnetic resonance (NMR) or
other biophysical methods [2,3]. Such screening typically
yields hits with weak activity in the high micromolar to
millimolar range. Although weak, fragment hits have high
'ligand efficiency' and, therefore, can be optimized by
'growing' the compound to bind into neighboring pockets of
the target protein or by connecting it to other fragment hits –
an approach that is discussed in a later section [4].
Furthermore, fragment screening hits can be used to
optimize leads found via more traditional methods, for
example, by connecting the fragments onto existing
The accuracy of structure-based computational methods is
affected by the quality and quantity of the available target
and ligand structural data. Structure-based virtual
screening, or molecular docking, is highly dependent on the
biological relevance of the existing 3-D structures of the
target. Solving the X-ray crystal structure of the target
co-crystallized with a number of different ligands is often
critical to the success of a modeling and structure-based
design effort. Consideration of target flexibility or subtle
changes in protein conformation upon ligand binding can be
crucial to the prediction of the docking mode of a ligand and
is reviewed below. In addition, determining new protein-
ligand complex structures periodically throughout the lead

Page 2
Lead optimization via high-throughput molecular docking Joseph-McCarthy et al 265
Figure 1. The structure-based lead optimization process.
Fragment screening
Virtual screening
screening of
corporate collections
Synthesis of new
compounds and
assays to
confirm hits
design of
improved lead
Structure of lead
bound to target
Structure of the
Biophysical techniques:
Protein biochemistry
X-ray crystallography
Nuclear magnetic resonance
Mass spectrometry
Steps that may involve the use of high-throughput docking techniques are highlighted with shadowed boxes.
optimization process allows the protein-binding site model
used for docking to be refined (if necessary) and provides a
check on the initial docking mode predictions. If the modeler
is confident that the original models can explain the majority
of the structure-activity relationships (SAR) observed via
biological testing, then solving these additional complex
structures may not be necessary. Furthermore, the
determination of protein-ligand structures can sometimes
prove difficult, even if the apo-structure is crystallized
A target structure that has been determined by 3-D NMR
techniques can be utilized for lead optimization; however,
this is less common than using an X-ray crystal structure,
because solving an NMR-determined protein structure can
be more time consuming and requires larger samples of
protein. NMR techniques are used for target determination
if the target is (i) sufficiently small (< 30 kDa), (ii) has
eluded structural determination via X-ray crystallographic
techniques, and (iii) is considered an important target.
However, it is more common that a homology model will be
generated based on a high resolution X-ray crystal structure
of a related protein, if one exists. For such homology models,
generally 20 to 30% sequence identity is required between
the desired target and the template protein sequence,
although there are no hard rules. For example, given low
overall sequence identity, higher identity in the active site
can be critical as can be the availability of other types of low
resolution structural data that provide constraints for the
target structure [5]. Gilson and co-workers carried out a
systematic study comparing the docking of a database of
'drug-like' molecules to an X-ray crystal structure and a set
of homology models for five drug targets [6]. The study
demonstrated that docking to target homology models can
result in significant enrichment of known actives in a ranked
hit list; however, the researchers often found similar
enrichments when docking directly to the templates for the
homology models themselves.
Annotated databases of 3-D structures of druggable binding
sites suitable for docking studies (eg, http://bioinfo-pharma. [7]) and comparative models for protein
sequences that are homologous to at least one existing 3-D
protein structure (eg, [8]) are
available on the Internet. Therefore, given the fact that the
number of 3-D macromolecular structures determined and
protein models generated continues to increase every year
[9], successful docking to surrogate protein structures and
homology models is also expected to increase in the future.
The success of a structure-based lead optimization effort also
depends, to a large extent, on close collaboration between
the modeler(s) and medicinal chemists developing the
ligand. Synthetic accessibility clearly needs to be considered
when designing modified leads, which is best accomplished
by the modeler and medicinal chemists together viewing
models of proposed compounds in 3-D. Tools such as
Benchware3D Explorer allow the modeler to prepare
annotated, labeled views of modeling results that can be
shared with medicinal chemists via email, and allows the
chemists to re-evaluate the models and structures at their
Desktop during the design of new synthetic targets. Existing
X-ray crystal structures of protein-ligand complexes are
viewed superimposed with docked poses for proposed
The structure-based lead optimization process can involve
simple energy minimization of a few molecules in the
protein-binding site or the docking of relatively large
combinatorial libraries of hundreds or even thousands of
molecules. The same docking methods that are used for the
virtual screening of large combinatorial libraries of millions

Page 3
266 Current Opinion in Drug Discovery & Development 2007 Vol 10 No 3
of compounds for lead generation are also often applied in
lead optimization. For lead optimization, however, false
negatives are generally considered to be a greater problem,
as fewer molecules are considered and therefore somewhat
lower throughput methods (which are expected to be more
accurate) are also employed. Because of the current
pressures to develop drugs more efficiently, lead
optimization often takes place under tight time constraints;
the rapid optimization of two series with a pre-development
decision in two years is typical for most projects and
therefore computational approaches that can aid in the
process are highly valued.
Molecular docking tools
Structure-based virtual screening involves docking a library
of proposed or existing small molecules into a target-
binding site to identify which molecules have a
complementary fit to the target binding site and are
therefore likely to bind to the target. When optimizing lead
compounds, the resulting binding modes and scores for
proposed compounds determined by docking experiments
are compared with those for the current best leads for the
target. Often the modeler will look through several highly
ranked poses (eg, 3 to 10) for each compound in the target
binding site and select the best scoring pose that maintains
the known position of the scaffold (or core of the molecule)
in the site. For kinase targets, for example, Perola reported
that requiring docked ligand poses form two hydrogen
bonds to the hinge region and that the hydrogen bonds to
fall into one of three preferred motifs dramatically reduced
the number of false positives in hit lists [10]. As such, both
sampling and scoring are important components of any
molecular docking approach (Figure 2). The objective
evaluation and comparison of different docking methods is,
however, difficult because each researcher may prepare their
binding sites and ligands differently, define the box
encompassing the binding site differently, or select 'expert
user', non-default parameters for a docking protocol.
Furthermore, different docking methods fail for different
types of systems and may enrich for different types of hits.
Despite the limitations to the docking experiments discussed
above, many research groups have tried to evaluate
commonly used docking methods. The overall conclusion of
a recent comparison of the DOCK, DOCKVISION, Glide and
GOLD methods for five protein targets was that all four
methods enriched hit rates compared with random
screening, but that the prioritization of known ligands is
both method- and target-dependent [11]. As a general-
purpose docking tool, Glide performed the best across the
set of five targets, however, DOCK, DOCKVISION and
GOLD each outperformed Glide on certain targets. More
Figure 2. Molecular docking schematic demonstrating possible hierarchical levels for scoring virtual ligands.
Molecules to
synthesize and test
ID 0000400
ID 0000043
ID 0001345
ID 0000022
ID 0000001
ID 0023452
Virtual library docked
(i) Shape
(ii) Electrostatics
(iii) Solvation
(A) Virtual library is docked into the target. (B) The virtual ligands are scored hierarchically according to, for example, (i) shape only,
(ii) electrostatics and (iii) solvation. The orientation and conformation of each ligand is sampled in the binding site and scored to produce
(C) The final ranked list of molecules for synthesis and testing.

Page 4
Lead optimization via high-throughput molecular docking Joseph-McCarthy et al 267
recently, Chen et al compared the FlexX, GOLD, Glide and
ICM docking methods for their ability to reproduce X-ray
protein-ligand complex structures and to enrich known
actives in hit lists [12]. This study went on to compare the
docking results to enrichments from some ligand-based
virtual screening experiments. For enriching known actives
in a hit list over a set of 12 diverse targets, the average
enrichment factors for ICM, Glide and ROCS (using a bound
conformation for the ligand from an X-ray protein-ligand
complex structure as the query) were greater than
4.6, while FlexX and GOLD gave enrichment factors of less
than the 3.5 obtained through ISIS 2-D similarity searching.
Warren et al tested 10 docking programs and 37 scoring
functions on eight protein targets, and also concluded that
no single program performed well for all targets [13]. These
results underline the need to have multiple docking tools
available, and, if possible, to test the ability of each method
to reproduce any existing protein-ligand X-ray crystal
structures prior to beginning a lead optimization or lead
generation effort.
In addition to comparing existing methods, there continue to
be reports of new or significantly improved docking
methods. For example, Thomsen et al have developed a new
docking approach that uses a guided differential evolution
algorithm with a modified piecewise linear potential (PLP),
and a re-ranking procedure that re-scores poses from
independent docking runs with a torsional and van der
Waals term added to the PLP [14]. Furthermore, Park et al
described a significantly improved automated version of
AutoDock [15].
Measures of accuracy
To an extent, the measures used to determine the accuracy
of docking methods depend on the overall goal of the
docking exercise. If high-throughput docking is used to
screen libraries and identify compounds for experimental
testing then the ability to separate actives from inactives or
to rank more actives higher relative to inactives is important.
In such cases, it is appropriate to use measures of
enrichment or ROC (receiver operating characteristic) curves
to evaluate different docking methods [16]. If the results of
docking experiments are being used for lead optimization
the determination of the true binding mode is important
[17]. The ability of a docking method and scoring scheme to
predict binding modes is usually evaluated by calculating
the symmetry-corrected RMSD between predicted and
actual binding modes for compounds with known crystal
structures. As such, the regeneration of a large set of
protein-ligand X-ray crystal structures has been used to
compare the performance of a number of scoring functions
[18]. Other methods of assessing the accuracy of docking
poses have been suggested, including a method developed
by Kroemer et al that concentrates on the interactions
between the ligand and the protein rather than just
deviations in the atomic positions [19].
As discussed, Chen et al examined the ability of docking
methods to enrich known actives in hit lists and to
accurately predict binding modes. They found that for both
tasks ICM and Glide performed particularly well, at least for
the test sets considered [12]. Previous studies by, for
example, Verdonk et al also highlighted the fact that to
achieve good enrichment in virtual screening experiments, it
is necessary to produce reliable binding modes [20]. Clearly,
when measuring the enrichment of actives in a screening
exercise the selection of test sets (containing known actives
and decoy compounds) is vitally important and has been the
focus of research by a number of groups [20,21••]. When
assessing the performance of docking methods for
compound optimization, the use of 'property-matched'
decoy sets is key because the goal is often to differentiate
between similar compounds, and to pick those most likely to
be active.
Scoring methods
The scoring functions used in molecular docking methods
can be divided into three catagories: (i) energy-based,
(ii) empirical, and (iii) knowledge-based. While virtual
screening of molecular databases is able to enrich hit
molecules in ranked lists of a relatively small size, in most
cases there is little correlation between the experimentally
determined binding affinity of the hits and the computed
score. Analysis of variance (ANOVA) techniques have
recently been used to quantify the discriminatory power of
scoring functions with respect to ligands and decoys, and
may facilitate the development of more accurate scoring
schemes [22]. In this section we discuss recent developments
in the field of scoring function design and use.
Energy-based scoring functions
Energy-based scoring functions approximate atomic
interactions between protein and ligand by including terms
known to be important in molecular recognition. In general,
they consist of the bonded and non-bonded terms common
to established molecular mechanics (MM) force fields
(eg, AMBER95 [23] and CHARMm22 [24]). The parameters
in these functions are derived from physical measurements,
and do not necessarily
include binding affinity
measurements, therefore the scoring functions are highly
transferable from small test systems to larger biological
systems. However, MM-based scoring functions do not
include solvation or entropic considerations and, as such,
calculate binding enthalpies as opposed to free energies.
Solvation and some entropic factors can be accounted for
implicitly by adding either a continuum solvent Poisson-
Boltzmann (PB) or a generalized Born (GB) term for polar
solvation, and a solvent-accessible surface area (SA) term for
nonpolar solvation. The resulting methodology is referred to
as MM-PBSA or MM-GBSA and approximates the free
energy of ligand binding. The MM-GBSA approach has been
shown to improve early enrichment rates in the virtual
screening of large compound databases [25] and to correctly
rank a series of congeneric kinase inhibitors [26••], while the
MM-PBSA approach was used to examine in detail the
binding affinities of a set of biotin analogs for avidin [27].
The linear interaction energy (LIE) method represents
a compromise
the more rigorous
computationally demanding free energy perturbation (FEP)
calculations and simple scoring functions. In contrast to FEP,
LIE uses only initial and final states of the binding event of a

Page 5
268 Current Opinion in Drug Discovery & Development 2007 Vol 10 No 3
single ligand to calculate binding free energies [28]. In a
recent study by Stjernschantz and co-workers, an automated
procedure for the straightforward use of the LIE method for
lead optimization is presented, and for three of the four
systems studied, the LIE method out-performed ten
different scoring functions [29].
Empirical scoring functions
Empirical scoring functions decompose the protein-ligand
binding affinity into a series of terms believed to be
important for binding free energy, with each term assigned a
weighting coefficient determined by a mathematical fit to
experimental binding data. Because of the reliance on a finite
sample of experimental data, empirical functions may
exhibit non-transferability issues, and it can be difficult to
know exactly what each term accounts for and to assess
where errors arise. As an example, the Glide XP 4.0 scoring
function includes both a water desolvation energy term
(a crude explicit water model) as well as terms that account
for specific protein-ligand structural motifs that provide
'exceptionally large contributions' to enhance binding
affinity (eg, hydrophobic enclosure, where a lipophilic
portion of a ligand is enclosed on opposite sides by
lipophilic protein atoms). The scoring function and
associated docking protocol have been developed to
reproduce experimental binding affinities for a set of 198
complexes and to yield significant database enrichments
against targets of pharmaceutical interest [30].
Knowledge-based scoring functions
Knowledge-based functions are derived from a statistical
analysis of the occurrence of atom-atom interactions in
known structures. The frequency of this occurrence can be
converted into a free energy term using a Boltzmann
distribution. Employing this method, the DrugScore
program was used to calculate potentials derived from
protein-ligand complexes [31]. A new variant of this
program, DrugScore
, has been developed based on data
generated from the crystal packing of small organic
molecules available through the Cambridge Structural
Database [32]. The highly resolved small molecule structures
provide relevant contact data across a better balanced
distribution of atom types found in drug-like molecules, and
produce potentials of superior statistical significance. The
original DrugScore and another commonly used knowledge-
based function, potentials of mean force (PMF), were
compared with a novel atom-atom potential derived from a
database of protein-ligand complexes – the Astex statistical
potential (ASP) [33]. ASP was used to construct a targeted
scoring function for cyclin-dependent kinase-2 that
produced significantly
improved enrichment rates
compared with the original ASP function. As might be
expected, it was demonstrated that the more structures used
in the construction of the targeted function the better the
overall results.
Consensus scoring
Consensus scoring methods can also be used to improve the
overall performance in docking methods [34,35]. Consensus
scoring methods have been applied to docking results by
generating a set of poses for each ligand using a particular
docking program, and then using multiple functions to re-
score the poses. Research using these methods is aimed at
establishing the performance of consensus scoring
techniques in various scenarios and determining the best
approach for combining scores [36,37]. Research on
consensus scoring in ligand-based screening has indicated
that much of the improvement from combining scores comes
from the fact that different methods have different
systematic errors [38]. Therefore, additional improvement
may be obtained by repeating the entire docking process
using multiple methods rather than simply re-scoring. The
GFscore method described by Betzi et al uses a non-linear
neural network to combine five scoring functions [39].
However, such methods require additional training sets or
computational expense, and some form of the original sum
rank method still appears to be the most practical solution
for real-life problems.
Lead optimization versus lead generation
High-throughput methods
High-throughput docking methods are routinely used both
for lead generation and lead optimization. For lead
generation it is necessary to screen large databases of up to
millions of compounds in a matter of days. While such
speeds are not required for lead optimization, it would
certainly be desirable if accurate results could be obtained in
seconds to assist the collaborative decision making between
modelers and chemists in designing molecules. Higher
throughput can be obtained by increasing computational
power, by using heuristics or scoring schemes that reduce
the time required to dock individual compounds, and by
reducing the number of compounds to be docked by
filtering (by, for example, pre-screening based on similarity
[40] or physical properties).
The docking process itself can be made faster through
reduced sampling by limiting the number of initial poses or
optimization steps or by restricting the poses based on
information known about the target. It is crucial to account for
ligand flexibility to screen a molecular database accurately,
and docking methods accomplish this either by flexing the
ligands during the docking stage or by pre-computing
conformers for each compound. Although the former method
requires much less computer disk space, the latter is
particularly useful when the same set of compounds (for
example, a corporate database) is screened against multiple
targets. The PhDock approach [41,42], for example,
significantly improves process speed by docking ensembles of
pre-computed conformers based on the largest 3-D
pharmacophore of each conformer and matching only ligand
pharmacophore points to target-derived pharmacophore
points. Further speedup is obtained by applying a tiered-
scoring approach, whereby the DOCK Contact score is used
as a fast filter of the entire database, followed by re-scoring of
poses for the top ranked molecules with a more physically
realistic function. Lorber et al have recently published a
method of pre-organizing similar conformer databases in a
hierarchical manner in order to consistently represent the
conformers and, furthermore, recognize and omit
incompatible conformations quickly without requiring a full

Page 6
Lead optimization via high-throughput molecular docking Joseph-McCarthy et al 269
docking run [43]. In addition, with conformationally
expanded databases, docking speed can be increased by
limiting the number of starting ligand conformations, and a
number of groups have investigated how this affects the
accuracy of the docking results [44-46]. In contrast, DOCK6.1
is just one example of a program that can be used to account
for ligand flexibility ad hoc during the docking process [47].
The Glide program also flexes the ligands as the docking
proceeds, and can be used with a tiered scoring approach in
a triage process [48,49].
Lower- to medium-throughput methods
Lower- to medium-throughput methods attempt to account
for solvation or protein flexibility in some way. Accounting
for solvation can be accomplished in a post-docking step by
saving multiple poses per ligand from a docking run and
then re-scoring the poses using a MM-GBSA or MM-PBSA
function. Verdonk et al have implemented a novel approach
in the GOLD program that allows explicit water molecules,
in particular those known to be crucial for binding, to switch
on and off and rotate during the docking process [50]. They
have used the approach, which applies a constant penalty to
the score to reward water displacement, to correctly predict
water-mediated protein-ligand complexes and water
Protein flexibility is in some ways more difficult than
solvation to address, and many research groups are
attempting to include at least some limited protein flexibility
in the docking process. Erickson et al examined the effect of
docking to the corresponding protein structure of a protein-
ligand complex, an average protein structure, and an apo
protein structure [51]. It was discovered, as predicted, that
docking accuracy falls off dramatically if one uses an
average protein or apo protein structure. Programs such as
Glide dock to a rigid protein but attempt, in the simplest
way, to account for some protein flexibility by using a
reduced van der Waals (vdw) potential. Sherman et al
developed an induced-fit docking (IFD) protocol [52••], in
which multiple receptor conformations are generated using
homology modeling software (PRIME) as starting points for
rigid receptor docking. In the PRIME approach, the protein
backbone is only minimized and not sampled extensively as
the side chain positions are. For the 21 test systems studied,
the IFD proctocol significantly improved the binding
predictions compared with those obtained when using a
rigid body model, and for 18 of the 21 cases the RMSD was
less than or equal to 1.8 Å. The induced-fit docking
approach described by Mizutani et al also uses a reduced
vdw potential grid for the initial docking phase, followed by
minimization of the protein-ligand complex [53].
Another approach for addressing protein flexibility that
continues to be extensively explored is docking to ensembles
of protein conformations (either structures or models). Ferrari
et al compared the use of a reduced vdw potential to docking
to an ensemble of protein structures [54]. They reported that
while the reduced vdw potential was always better than the
full vdw potential at identifying known ligands when
docking to a single protein structure, the reduced vdw
potential was worse than the full potential when docking to
multiple protein structures for a given target. The FlexE
approach creates an ensemble by superimposing the
structures for a given target, merging the similar parts, and
explicitly taking into account and allowing combinatorial
recombination of the varying parts of the protein. In the
original paper describing FlexE, the approach was validated
on a set of ten protein ensembles [55]. However, Polgár and
Keserü more recently compared the use of FlexE with FlexX
and FlexX-Pharm. They reported that, when docking ligands
to c-Jun N-terminal kinase-3 and
-secretase, the FlexE
approach was not capable of predicting protein loop
movements. Furthermore, even when using FlexE to predict
side chain flexibility it did not outperform FlexX which
maintains a rigid protein [56]. However, they did report that
FlexE was useful for rapidly docking to proteins with
different side chain protonation states. Huang and Zou
presented an alternative ensemble docking algorithm that
allows the scoring and global optimization procedure of their
docking approach to automatically select an optimal protein
structure from an ensemble for each ligand [57]. Furthermore,
Cavasotto et al presented a novel algorithm for generating
alternative protein conformations via normal mode analysis
by perturbing the starting structure along a combination of
relevant modes. The resulting protein conformations were
then used for ensemble docking with the ICM program [58].
Finally, protein flexibility is readily incorporated with
Monte Carlo or molecular dynamics simulation approaches
that allow the protein and ligand to relax simultaneously.
Researchers at Wyeth have used rapid Monte Carlo docking
successfully on a number of lead optimization projects, as
described below in the case studies section of this review.
Combinatorial library enumeration and docking
One of the first steps towards optimizing a series of lead
compounds is to determine which compounds can be easily
synthesized using an existing synthetic route. This step
involves examining a list of commercially available reagents
of a given class, which could be substituted at a relevant
position on the lead to generate potentially more active
compounds. In some cases, depending on whether the
synthesis is convergent or divergent and what the level of
automation is, it may be possible to substitute more than one
position leading to a huge number of synthetically accessible
compounds [59].
With improvements in docking speed it is often practical to
enumerate large virtual libraries and score every compound.
CombiGlide from Schrodinger LLC, is a relatively new tool
for the design of focused combinatorial libraries that can
rapidly screen very large numbers of compounds and
eliminate unpromising ones at an early stage. The authors'
group successfully used CombiGlide to optimize a series to
increase the in vitro activity of a lead compound by 20- to
30-fold in a single step [Feyfant E, unpublished data]. Given
an experimentally determined or a predicted binding mode
for a scaffold in a target-binding site, CombiGlide allows the
user to attach reagents based on desired chemistry. In this
study, the CombiGlide diverse side chain database, which
consists of only 821 reagents, was screened. Substitutions
occurred at only one position on the lead core, and in an
automated process the combinatorial library was fully

Page 7
270 Current Opinion in Drug Discovery & Development 2007 Vol 10 No 3
enumerated; absorption, distribution, metabolism and
excretion (ADME) properties were calculated using QikProp
[60], and the library was docked. Compounds that scored
well, had desired ADME properties and passed a visual
inspection were then proposed for synthesis.
When using more complicated combinatorial libraries,
docking scores can be combined with other criteria into a
fitness score to select compounds for synthesis that optimize
the overall fitness of the library. In a number of publications,
reagents have been either filtered and pruned, or adapted
recursively to optimize the final library while maintaining
the combinatorial nature of the synthesis (see, for example,
[61] and [62]).
Fragment or 'small lead' optimization
Fragment screening is another, increasingly common
approach for lead generation and optimization. Fragments
are defined as small molecules typically with molecular
weight (MW) of less than 300 Da. Because the size of the
chemical space is proportional to the size limit of the
molecules considered, covering chemical space is more
manageable with fragments than with drug-sized molecules.
In addition, as mentioned in the introduction, the ligand
efficiency of a fragment may make it more amenable to
optimization than a drug-like molecule. However, the
challenges of optimizing fragments as leads are somewhat
different to those for complete molecules. Furthermore,
while optimization of a fragment or selection of fragments,
with a view to connecting them, may involve high-
throughput docking, the techniques often need to be
modified or augmented with additional computational
approaches (Figure 3). A number of successful examples of
using fragment-based approaches to develop a drug
candidate or a new lead series in a pharmaceutical project
have been reported (see for example [63]).
Figure 3. A fragment-based screening paradigm.
Target binding site
+ bound fragments
Fragment-based optimization
Connect fragments
Fragment as new scaffold
Library enumeration
Target binding site + lead
Modify existing leads
The diagram demonstrates how optimized ligands can be constructed by (A) connecting bound fragments, (B) optimizing existing leads by
adding fragments, or (C) selecting a bound fragment as a new ligand scaffold.

Page 8
Lead optimization via high-throughput molecular docking Joseph-McCarthy et al 271
Howard et al at Astex Therapeutics Ltd developed a novel
thrombin inhibitor using a fragment-linking approach
[64••]. A thrombin-focused fragment library was designed
by virtual screening of a subset of an in-house library
against several conformations of thrombin. Based on the
molecular docking results, 80 fragments were selected for
screening by X-ray crystallography following Astex's
pyramid screening paradigm [65]. Among the binders
detected by X-ray crystallography screening, the authors
chose three neutral fragments for optimization – two that
bound to thrombin in the S1 pocket (IC
= 330 µM and
> 1 mM, respectively) and one that bound in the S2-S4
pocket (IC
= 100 µM). The overlay of the X-ray crystal
structures revealed a clear opportunity to link the S2-S4
binder with the S1 binders. First, a small library of analogs
of the S2-S4 pocket binder was synthesized, and a more
potent compound was selected (IC
= 12 µM). Then, each of
the S1 pocket-binders was linked to the new S2-S4 binder,
respectively. The resulting compounds showed 50- to
200-fold increased potency (with IC
values of 220 nM and
1.4 nM) and selectivity versus trypsin.
Card et al discovered a new family of phosphodiesterase
(PDE) inhibitors using a similar approach [66]. A library of
20,000 compounds with molecular weight ranging from
125 to 350 Da was screened by biophysical assay against
several PDEs, and 316 binders were identified. After
attempting to co-crystallize all the binders with PDE, one of
the low-affinity hits (PDE4D IC
= 82
M; MW 168 Da) was
selected as a possible scaffold (a scaffold, as defined by the
authors, is a compound that forms key interactions with the
receptor and is expected to retain its binding mode upon
minor substitution). A small number of close analogs were
tested and the most potent analog (IC
= 270 nM) was
co-crystallized with the protein confirming the expected
binding mode. A virtual combinatorial library around this
new scaffold was enumerated, according to a synthetic
schema, docked into the binding site, and scored using an
MM-PBSA method. Using this approach, combining
experimental fragment screening with molecular docking to
grow an initial fragment, the authors were ultimately able
to design a relatively low MW (289 Da), high potency
= 30 nM) inhibitor.
Case studies
The literature is full of reports of successful cases of
structure-based lead optimization (see for example [67-70])
and, as such, a comprehensive review is impractical here.
Examples from Wyeth include research on the metabolic
disease target phosphotyrosine phosphatase (PTP)1B [71,72],
the inflammation targets tumor necrosis factor-α-converting
enzyme [73] and cytosolic-phospholipase A2α [74], and the
infectious disease target acyl carrier protein synthase [75]. In
the case of PTP1B, structure-guided lead optimization led to
the rapid optimization of a lead compound with an IC
value of 230 µM, through a number of iterations, resulting in
a 4-nM compound in approximately 9 months [Joseph-
McCarthy D, unpublished data]. The inclusion of some
limited protein flexibility was critical to the success of the
modeling. Side chains in the binding site were allowed to
undergo constrained movement during rapid Monte Carlo
docking [76], and the binding site model including the
choice of flexible residues was refined throughout the
process as new protein-ligand X-ray crystal structures were
Future advances in the area of structure-based lead
optimization will likely involve the development of better
scoring functions and the inclusion of protein flexibility into
more high-throughput docking approaches. Rigorous
theoretical work and increases in computer power continue,
for example, to enable ever more accurate approximations of
solvation and entropic effects [77-80]. Furthermore,
increasingly accurate ligand charge states are being
determined through the use of quantum mechanics in the
presence of continuum solvent [81], and by directly
optimizing ligand charges using sensitivity analysis [82].
As another example of research that may impact lead
optimization methods, Hao has presented a quantum
mechanics approach for calculating hydrogen bond
strengths for drug molecules [83]. Finally, over the next five
years, fragment screening, both experimental and in silico,
will probably play a greater role in the drug discovery
process. As a result, methods that are more finely tuned for
the docking and scoring of low MW fragment molecules as
very weak inhibitors will almost certainly emerge. Overall, it
is anticipated that more accurate, higher-throughput
molecular docking will continue to play an increasing role in
the lead optimization process.
of outstanding interest
of special interest
Wunberg T, Hendrix M, Hillisch A, Lobell M, Meier H, Schmeck C, Wild
H, Hinzen B: Improving the hit-to-lead process: Data-driven
assessment of drug-like and lead-like screening hits. Drug Disc
Today (2006) 11(3-4):175-180.
Carr RA, Congreve M, Murray CW, Rees DC: Fragment-based lead
discovery: Leads by design. Drug Disc Today (2005) 10(14):987-992.
Ciulli A, Williams G, Smith AG, Blundell TL, Abell C: Probing hot spots
at protein-ligand binding sites: A fragment-based approach using
biophysical methods. J Med Chem (2006) 49(16):4992-5000.
Cele AZ, Metz JT: Ligand efficiency indices as guideposts for drug
discovery. Drug Disc Today (2005) 10(7):464-469.
Reviews the various ligand efficiency definitions and indices, and discusses
their utility.
Topf M, Baker ML, Marti-Renom MA, Chiu W, Sali A: Refinement of
protein structures by iterative comparative modeling and cryoEM
density fitting. J Mol Biol (2006) 357(5):1655-1668.
Kairys V, Fernandes MX, Gilson MK: Screening drug-like compounds
by docking to homology models: A systematic study. J Chem Inf
Model (2006) 46(1):365-379.
While in general, docking to homology models can enrich for known actives,
standard measures of similarity between the template and target protein do
not correlate with enrichment rates obtained using a given homology mode.
This paper discusses and examines these effects on carboxypeptidase A,
coagulation Factor Xa, peroxisome proliferator-activated receptor
, cyclin
dependent kinase 2 and acetylcholinesterase.
Kellenberger E, Muller P, Schalon C, Bret G, Foata N, Rognan D:
sc-PDB: An annotated database of druggable binding sites from
the protein data bank. J Chem Inf Model (2006) 46(2):717-727.
Annotations for 6415 binding sites include protein name, function, source,
domain and mutations, ligand name and structure.

Page 9
272 Current Opinion in Drug Discovery & Development 2007 Vol 10 No 3
Pieper U, Eswar N, Davis FP, Braberg H, Madhusudhan MS, Rossi A,
Marti-Renom M, Karchin R, Webb BM, Eramian D, Shen MY et al:
MODBASE: A database of annotated comparative protein structure
models and associated resources. Nucleic Acids Res (2006)
34(Database Issue):D291-D295.
Berman HM, Burley SK, Chiu W, Sali A, Adzhubei A, Bourne PE, Bryant
SH, Dunbrack RL Jr, Fidelis K, Frank J, Godzik A et al: Outcome of a
workshop on archiving structural
macromolecules. Structure (2006) 14(8):1211-1217.
10. Perola E: Minimizing false positives in kinase virtual screens.
Proteins (2006) 64(2):422-435.
11. Cummings MD, DesJarlais RL, Gibbs AC, Mohan V, Jaeger EP:
Comparison of automated docking programs as virtual screening
tools. J Med Chem (2005) 48(4):962-976.
12. Chen H, Lyne PD, Giordanetto F, Lovell T, Li J: On evaluating
molecular-docking methods for pose prediction and enrichment
factors. J Chem Inf Model (2006) 46(1):401-415.
This paper assesses the virtual screening and ligand docking capabilities of
four docking methods (FLexX, GOLD, GLIDE and ICM). On a target-by-target
basis, ICM performed best for 8 of the 12 targets, Glide for 4 and ROCS for 2.
It is noted that binding site preparation and control parameter settings may
affect the results.
13. Warren GL, Andrews CW, Capelli AM, Clarke B, LaLonde J, Lambert
MH, Lindvall M, Nevins N, Semus SF, Senger S, Tedesco G et al:
A critical assessment of docking programs and scoring functions.
J Med Chem (2006) 49(20):5912-5931.
14. Thomsen R, Christensen MH: MolDock: A new technique for high-
accuracy molecular docking. J Med Chem (2006) 49(11):3315-3321.
15. Park H, Lee J, Lee S: Critical assessment of the automated
AutoDock as a new docking tool for virtual screening. Proteins
(2006) 65(3):549-554.
16. Triballeau N, Acher F, Brabet I, Pin JP, Bertrand HO: Virtual screening
workflow development guided by the "receiver operating
characteristic" curve approach. Application to high-throughput
docking on metabotropic glutamate receptor subtype 4. J Med
Chem (2005) 48(7):2534-2547.
A good description of the use of ROC curves to guide high-throughput
17. Jain AN: Virtual screening in lead discovery and optimization. Curr
Opin Drug Discovery Dev (2004) 7(4):396-403.
18. Wang R, Lu Y, Fang X, Wang S: An extensive test of 14 scoring
functions using the PDBbind refined set of 800 protein-ligand
complexes. J Chem Inf Comput Sci (2004) 44(6):2114-2125.
19. Kroemer RT, Vulpetti A, McDonald JJ, Rohrer DC, Trosset JY,
Giordanetto F, Cotesta S, McMartin C, Kihlen M, Stouten PF:
Assessment of docking poses: Interactions-based accuracy
classification (IBAC) versus crystal structure deviations. J Chem Inf
Comput Sci (2004) 44(3):871-881.
20. Verdonk ML, Berdini V, Hartshorn MJ, Mooij WT, Murray CW, Taylor
RD, Watson P: Virtual screening using protein-ligand docking:
Avoiding artificial enrichment. J Chem Inf Comput Sci (2004)
21. Huang N, Shoichet BK, Irwin JJ: Benchmarking sets for molecular
docking. J Med Chem (2006) 49(23):6789-6801.
•• A standard benchmarking set of compounds called the directory of useful
decoys (DUD). In this set, the decoy compounds (those that are either known
or assumed to be inactive against a particular target) are physically similar yet
topologically distinct to the known actives.
22. Seifert MH: Assessing the discriminatory power of scoring
functions for virtual screening. J Chem Inf Model (2006) 46(3):1456-
23. Cornell WD, Cieplak P, Bayly CI, Gould IR, Merz KM, Ferguson DM,
Spellmeyer DC, Fox T, Caldwell JW, Kollman PA: A 2nd generation
force-field for the simulation of proteins, nucleic-acids, and
organic-molecules. J Am Chem Soc (1995) 117(19):5179-5197.
24. MacKerell AD Jr, Bashford D, Bellott M, Dunbrack RL Jr, Evanseck JD,
Field MJ, Fischer S, Gao J, Guo H, Ha S, Joseph-McCarthy D et al:
All-atom empirical potential for molecular modeling and dynamics
studies of proteins. J Phys Chem B (1998) 102(18):3586-3616.
25. Huang N, Kalyanaraman C, Irwin JJ, Jacobson MP: Physics-based
scoring of protein-ligand complexes: Enrichment of known
inhibitors in large-scale virtual screening. J Chem Inf Model (2006)
26. Lyne PD, Lamb ML, Saeh JC: Accurate prediction of the relative
potencies of members of a series of kinase inhibitors using
molecular docking and MM-GBSA scoring. J Med Chem (2006)
•• An MM-GBSA function is used in a post-docking step to 're-score' docked
poses of each ligand.
27. Weis A, Katebzadeh K, Soderhjelm P, Nilsson I, Ryde U: Ligand
affinities predicted with the MM/PBSA method: Dependence on the
simulation method and the force field. J Med Chem (2006)
This investigation showed there was little dependence on choice of the force
field when using the MM-PBSA method to predict ligand affinities. However,
the mixing of force fields using, for example, one force field for molecular
dynamics simulations and another for the MM-PBSA energy calculations is
not recommended.
28. Aqvist J, Medina C, Samuelsson JE: A new method for predicting
binding affinity in computer-aided drug design. Protein Eng (1994)
29. Stjernschantz E, Marelius J, Medina C, Jacobsson M, Vermeulen NP,
Oostenbrink C: Are automated molecular dynamics simulations and
binding free energy
calculations realistic tools in
optimization? An evaluation of the linear interaction energy (LIE)
method. J Chem Inf Model (2006) 46(5):1972-1983.
30. Friesner RA, Murphy RB, Repasky MP, Frye LL, Greenwood JR,
Halgren TA, Sanschagrin PC, Mainz DT: Extra precision glide:
Docking and scoring incorporating a model of hydrophobic
enclosure for protein-ligand complexes. J Med Chem (2006)
Gohlke H, Hendlich M, Klebe G: Knowledge-based scoring function to
predict protein-ligand interactions. J Mol Biol (2000) 295(2):337-356.
32. Velec HF, Gohlke H, Klebe G: DrugScore(CSD)-knowledge-based
scoring function derived from small molecule crystal data with
superior recognition rate of near-native ligand poses and better
affinity prediction. J Med Chem (2005) 48(20):6296-6303.
33. Mooij WT, Verdonk ML: General and targeted statistical potentials
for protein-ligand interactions. Proteins (2005) 61(2):272-287.
34. Charifson PS, Corkery JJ, Murcko MA, Walters WP: Consensus
scoring: A method for obtaining improved hit rates from docking
databases of three-dimensional structures into proteins. J Med
Chem (1999) 42(25):5100-5109.
35. Stahl M, Rarey M: Detailed analysis of scoring functions for virtual
screening. J Med Chem (2001) 44(7):1035-1042.
36. Oda A, Tsuchida K, Takakura T, Yamaotsu N, Hirono S: Comparison
of consensus scoring strategies for evaluating computational
models of protein-ligand complexes. J Chem Inf Model (2006)
37. Yang JM, Chen YF, Shen TW, Kristal BS, Hsu DF: Consensus scoring
criteria for improving enrichment in virtual screening. J Chem Inf
Model (2005) 45(4):1134-1146.
38. Baber JC, Shirley WA, Gao Y, Feher M: The use of consensus
scoring in ligand-based virtual screening. J Chem Inf Model (2006)
39. Betzi S, Suhre K, Chetrit B, Guerlesquin F, Morelli X: GFscore:
A general nonlinear consensus scoring function for high-
throughput docking. J Chem Inf Model (2006) 46(4):1704-1712.
40. Vidal D, Thormann M, Pons M: A novel search engine for virtual
screening of very large databases. J Chem Inf Model (2006)
41. Joseph-McCarthy D, McFadyen IJ, Zou J, Walker G, Alvarez JC:
Pharmacophore-based molecular docking: A practical guide. In:
Virtual Screening in Drug Discovery. Alvarez JC, Shoichet B (Eds), CRC
Press, Boca Raton, FL, USA (2004).

Page 10
Lead optimization via high-throughput molecular docking Joseph-McCarthy et al 273
42. Joseph-McCarthy D, Thomas BE 4th, Belmarsh M, Moustakas D,
Alvarez JC: Pharmacophore-based molecular docking to account
for ligand flexibility. Proteins (2003) 51(2):172-188.
43. Lorber DM, Shoichet BK: Hierarchical docking of databases of
multiple ligand conformations. Curr Top Med Chem (2005) 5(8):739-
44. Kirchmair J, Laggner C, Wolber G, Langer T: Comparative analysis of
protein-bound ligand conformations with respect to Catalyst's
conformational space subsampling algorithms. Chem Inf Model
(2005) 45(2):422-430.
45. Knox AJ, Meegan MJ, Carta G, Lloyd DG: Considerations in
compound database preparation-"hidden" impact on virtual
screening results. J Chem Inf Model (2005) 45(6):1908-1919.
46. Yoon S, Welsh WJ: Identification of a minimal subset of receptor
conformations for improved multiple conformation docking and
two-step scoring. J Chem Inf Comput Sci (2004) 44(1):88-96.
47. Moustakas DT, Lang PT, Pegg S, Pettersen E, Kuntz ID, Brooijmans N,
Rizzo RC: Development and validation of a modular, extensible
docking program: DOCK 5. J Comput Aided Mol Des (2006) 20(10-
48. Friesner RA, Banks JL, Murphy RB, Halgren TA, Klicic JJ, Mainz DT,
Repasky MP, Knoll EH, Shelley M, Perry JK, Shaw DE et al: Glide:
A new approach for rapid, accurate docking and scoring. 1. Method
and assessment of docking accuracy. J Med Chem (2004)
49. Halgren TA, Murphy RB, Friesner RA, Beard HS, Frye LL, Pollard WT,
Banks JL: Glide: a new approach for rapid, accurate docking and
scoring. 2. Enrichment factors in database screening. J Med Chem
(2004) 47(7)1750-1759.
50. Verdonk ML, Chessari G, Cole JC, Hartshorn MJ, Murray CW, Nissink
JW, Taylor RD, Taylor R: Modeling water molecules in protein-ligand
docking using GOLD. J Med Chem (2005) 48(20):6504-6515.
51. Erickson JA, Jalaie M, Robertson DH, Lewis RA, Vieth M: Lessons in
molecular recognition: The effects of ligand and protein flexibility
on molecular docking accuracy. J Med Chem (2004) 47(1):45-55.
52. Sherman W, Day T, Jacobson MP, Friesner RA, Farid R: Novel
procedure for modeling ligand/receptor induced fit effects. J Med
Chem (2006) 49(2):534-553.
•• The paper describes a three-step process in which (i) the ligand is first
docked into the original target structure using Glide, and highly flexible
residues mutated to Ala; (ii) the 20 best poses are imported into PRIME, and
with wild-type side chains present the system is minimized; (iii) the 20 new
receptor conformations are used as rigid receptors for a new set of docking
53. Mizutani MY, Takamatsu Y, Ichinose T, Nakamura K, Itai A: Effective
handling of induced-fit motion in flexible docking. Proteins (2006)
54. Ferrari AM, Wei, BO, Costantino L, Shoichet BK: Soft Docking and
Multiple Receptor Conformations in Virtual Screening. J Med Chem
(2004) 47(12):5076-5084.
55. Claussen H, Buning C, Rarey M, Lengauer T: FlexE: Efficient
molecular docking considering protein structure variations. J Mol
Biol (2001) 308(2):377-395.
56. Polgar T, Keseru GM: Ensemble docking into flexible active sites.
Critical evaluation of FlexE against JNK-3 and β-secretase. J Chem
Inf Model (2006) 46(4):1795-1805.
57. Huang S-Y, Zou X: Ensemble docking of multiple protein structures:
Considering protein structural variations in molecular docking.
Proteins (2007) 66(2):399-421.
58. Cavasotto CN, Kovacs JA, Abagyan RA: Representing receptor
flexibility in ligand docking through relevant normal modes. J Am
Chem Soc (2005) 127(26):9632-9640.
59. Edwards PJ, Allart B, Andrews MJ, Clase JA, Menet C: Expediting
drug discovery: Recent advances in fast medicinal chemistry
optimization of hits and leads. Curr Opin Drug Discovery Dev (2006)
A review of recent advances in high-throughput synthesis techniques and
their use in optimization.
Duffy EM, Jorgensen WL: Prediction of properties from simulations:
Free energies of solvation in hexadecane, octanol, and water. J Am
Chem Soc (2000) 122(12):2878-2888.
61. Le Bailly de Tilleghem C, Beck B, Boulanger B, Govaerts B: A fast
exchange algorithm for designing focused libraries in lead
optimization. J Chem Inf Model (2005) 45(3):758-767.
62. Truchon JF, Bayly CI: GLARE: A new approach for filtering large
reagent lists in combinatorial library design using product
properties. J Chem Inf Model (2006) 46(4):1536-1548.
63. Schneider G, Fechner U: Computer-based de novo design of drug-
like molecules. Nat Rev Drug Disc (2005) 4(8):649-663.
64. Howard N, Abell C, Blakemore W, Chessari G, Congreve M, Howard S,
Jhoti H, Murray CW, Seavers LCA, van Montfort RL: Application of
fragment screening and fragment linking to the discovery of novel
thrombin inhibitors. J Med Chem (2006) 49(4):1346-1355.
•• Although an inhibitor with an IC
value of 1.4 nM was designed, the
authors note that they did not achieve the optimal ligand efficiency
theoretically possible, and they attribute this to the fact that the optimized
compounds did not fully retain the interactions of the initial fragment.
65. Hartshorn MJ, Murray CW, Cleasby A, Frederickson M, Tickle IJ, Jhoti
H: Fragment-based lead discovery using X-ray crystallography.
J Med Chem (2005) 48(2):403-413.
66. Card GL, Blasdel L, England BP, Zhang C, Suzuki Y, Gillette S, Fong D,
Ibrahim PN, Artis DR, Bollag G, Milburn MV et al: A family of
phosphodiesterase inhibitors discovered by cocrystallography and
scaffold-based drug design. Nat Biotechnol (2005) 23(2):201-207.
67. Ghosh AK, Sridhar PR, Leshchenko S, Hussain AK, Li J, Kovalevsky
AY, Walters DE, Wedekind JE, Grum-Tokars V, Das D, Koh Y et al:
Structure-based design of novel HIV-1 protease inhibitors to
combat drug resistance. J Med Chem (2006) 49(17):5252-5261.
68. Krovat EM, Fruhwirth KH, Langer T: Pharmacophore identification,
in silico screening, and virtual library design for inhibitors of the
human factor X
. J Chem Inf Model (2005) 45(1):146-159.
69. Lu IL, Huang CF, Peng YH, Lin YT, Hsieh HP, Chen CT, Lien TW, Lee
HJ, Mahindroo N, Prakash E, Yueh A et al: Structure-based drug
design of a novel family of PPARγ partial agonists: Virtual
screening, X-ray crystallography, and in vitro/in vivo biological
activities. J Med Chem (2006) 49(9):2703-2712.
70. Trosset JY, Dalvit C, Knapp S, Fasolini M, Veronesi M, Mantegani S,
Gianellini LM, Catana C, Sundstrom M, Stouten PF, Moll JK: Inhibition
of protein-protein interactions: The discovery of druglike β-catenin
inhibitors by combining virtual and biophysical screening. Proteins
(2006) 64(1):60-67.
Moretto AF, Kirincich SJ, Xu WX, Smith MJ, Wan ZK, Wilson DP,
Follows BC, Binnun E, Joseph-McCarthy D, Foreman K, Erbe DV et al:
Bicyclic and tricyclic thiophenes as protein tyrosine phosphatase
1B inhibitors. Bioorg Med Chem (2006) 14(7):2162-2177.
72. Wan ZK, Lee J, Xu W, Erbe DV, Joseph-McCarthy D, Follows BC,
Zhang YL: Monocyclic thiophenes as protein tyrosine phosphatase
1B inhibitors: Capturing interactions with Asp
. Bioorg Med Chem
Lett (2006) 16(18):4941-4945.
73. Condon JS, Joseph-McCarthy D, Levin JI, Lombart H-G, Lovering FE,
Sun L, Wang W, Xu W, Zhang Y: Identification of potent and selective
TACE inhibitors via the S1 pocket. Bioorg Med Chem Lett (2007)
74. Gopalsamy A, Yang H, Ellingboe JW, McKew JC, Tam S, Joseph-
McCarthy D, Zhang W, Shen M, Clark JD: 1,2,4-Oxadiazolidin-3,5-
diones and 1,3,5-triazin-2,4,6-triones as cytosolic phospholipase
A(2)α inhibitors. Bioorg Med Chem Lett (2006) 16(11):2978-2981.
75. Joseph-McCarthy D, Parris K, Huang A, Failli A, Quagliato D, Dushin
EG, Novikova E, Severina E, Tuckman M, Petersen PJ, Dean C et al:
Use of structure-based drug design approaches to obtain novel
anthranilic acid acyl carrier protein synthase inhibitors. J Med
Chem (2005) 48(25):7960-7969.
76. McMartin C, Bohacek RS: QXP: Powerful, rapid computer algorithms
for structure-based drug design. J Comput Aided Mol Des (1997)

Page 11
274 Current Opinion in Drug Discovery & Development 2007 Vol 10 No 3
77. Michel J, Taylor RD, Essex JW: Efficient generalized born models for
Monte Carlo simulations. J Chem Theory Computat (2006) 2(3):
78. Lu B, Cheng X, Huang J, McCammon JA: Order N algorithm for
computation of electrostatic interactions in biomolecular systems.
Proc Natl Acad Sci (2006) 103(51):19314-19319.
79. Carlsson J, Aqvist J: Calculations of solute and solvent entropies
from molecular dynamics simulations. Phys Chem Chem Phys
(2006) 8(46):5385-5395.
80. Salaniwal S, Manas ES, Alvarez JC, Unwalla RJ: Critical evaluation of
methods to incorporate entropy loss upon binding in high-
throughput docking. Proteins (2007) 66(2):422-435.
81. Grater F, Schwarzl SM, Dejaegere A, Fischer S, Smith JC:
Protein/ligand binding free energies calculated with quantum
mechanics/molecular mechanics.
J Phys
B (2005)
82. Gilson MK: Sensitivity analysis and charge-optimization for flexible
ligands: Applicability to lead optimization. J Chem Theory Comput
(2006) 2(2):259-270.
83. Hao MH: Theoretical calculation of hydrogen-bonding strength for
drug molecules. J Chem Theory Comput (2006) 2(3):863-872.