You are on page 1of 22

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/232740361

GAMESS as a Free Quantum-Mechanical Platform for Drug Research.

Article  in  Current topics in medicinal chemistry · October 2012


DOI: 10.2174/156802612804910269 · Source: PubMed

CITATIONS READS
58 1,462

4 authors, including:

Yuri Alexeev Michael P Mazanetz


Argonne National Laboratory NovaData Solutions Ltd, UK
79 PUBLICATIONS   760 CITATIONS    23 PUBLICATIONS   643 CITATIONS   

SEE PROFILE SEE PROFILE

Osamu Ichihara
Schrödinger Inc.
68 PUBLICATIONS   1,997 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

HSP90 inhibitors View project

Quantum Computing View project

All content following this page was uploaded by Yuri Alexeev on 20 January 2014.

The user has requested enhancement of the downloaded file.


Send Orders of Reprints at bspsaif@emirates.net.ae
Current Topics in Medicinal Chemistry, 2012, 12, 000-000 1

GAMESS As a Free Quantum-Mechanical Platform for Drug Research

Yuri Alexeev1, Michael P. Mazanetz2, Osamu Ichihara2 and Dmitri G. Fedorov3,*

1
Argonne Leadership Computing Facility, Argonne National Laboratory, 9700 South Cass Avenue, Building 240, Ar-
gonne, IL 60439, USA; 2Evotec (UK) limited, 114 Milton Park, Abingdon, Oxfordshire, OX14 4SA, UK; 3NRI, National
Institute of Advanced Industrial Science and Technology, Central 2, Umezono 1-1-1, Tsukuba 305-8568, Japan

Abstract: Driven by a steady improvement of computational hardware and significant progress in ab initio method devel-
opment, quantum-mechanical approaches can now be applied to large biochemical systems and drug design. We review
the methods implemented in GAMESS, which are suitable to calculate large biochemical systems. An emphasis is put on
the fragment molecular orbital method (FMO) and quantum mechanics interfaced with molecular mechanics (QM/MM).
The use of FMO in the protein-ligand binding, structure-activity relationship (SAR) studies, fragment- and structure-based
drug design (FBDD/SBDD) is discussed in detail.
Keywords: Quantum chemistry, fragment molecular orbital, drug design, ab initio, GAMESS, FMO, QM/MM, FBDD, SBDD.

1. INTRODUCTION try, is quickly disappearing. We also note that overall QM


calculations are more standard and transparent than for ex-
Molecular mechanics (MM) has been very successfully
ample, force fields.
applied to many systems in biochemistry [1]. Many elaborate
force fields such as CHARMM [2] or AMBER [3, 4] have The other factor of time scale, that is, the need to take
been developed and highly tuned to the studies of biochemi- into account the entropy and consider dynamic aspects of
cal systems. The traditional force fields offer high computa- biochemical process on a realistic time scale, still remains a
tional efficiency, however, the parameterization and the sim- major problem. It should be said here that while for force
plistic models have very considerable limitations. Two main fields one can get away with the dubious idea of simulating
drawbacks of the traditional force fields, the neglect of the processes at an unphysically high temperature to accelerate
polarization and charge transfer (CT), have been the subject them, the same will not work with QM, because bonds are
of developing a new generation of force fields [5-10]. Fre- not fixed by springs and will easily break as temperature
quently, parameters for force fields are generated from quan- increases.
tum-mechanical (QM) calculations of model systems. The The use of QM in drug research [11] is one of the most
question arises to the transferability of those parameters, exciting developments in the recent years which has the po-
which in the language of physics can be formulated as the tential to revolutionize this field. QM methods have impor-
importance of many-body effects in describing physical in- tant advantages over MM approaches. Firstly, QM has no
teractions. Importantly, the studies of chemical reactions implicit parameterization and relies only on the use of well-
with bond breaking are typically conducted with QM ap- known physical constants such as the velocity of light and
proaches. the masses and charges of atoms. The molecular properties
The reason why relatively few biochemical scientists and geometries are obtained by solving the differential
have taken advantage of QM approaches lies in several fac- Schrödinger equation [12-14]. Unfortunately, the equation
tors. One is their large computation cost. However, driven by can be solved directly only for only very small systems [15,
both the remarkable progress in QM method development 16].
and the amazing pace in computational hardware improve- A number of QM methods based on various models and
ment, with the revolutionary advent of multicore CPUs (a assumptions on the structure of the wave function have been
single node with 48 cores is readily available) and GPUs developed. Density functional theory (DFT) [17] owes its
(with hundreds of computing units), one can now routinely popularity to relatively low cost and a fairly good accuracy,
perform calculations of systems containing hundreds and and systems of biochemical size are becoming accessible
thousands of atoms. The other factor is the difficulty in set- [18]. Recently developed DFT methods deliver bond lengths
ting up and running the calculations. Many excellent graphi- for organic molecules within 0.02 Å and an absolute energy
cal packages enable users to employ QM methods with rela- error within 3 kcal/mol [14] in comparison to experiment.
tive ease. The aura of being too slow and difficult for the However, such high accuracy comes at a hefty computer
non-initiated, which has long loomed over quantum chemis- time price. Until recently, only fairly small systems (~100
atoms) could be computed because the cost of calculations
*Address correspondence to this author at NRI, National Institute of Ad- scales at least cubically with respect to the system size.
vanced Industrial Science and Technology, Central 2, Umezono 1-1-1,
Tsukuba 305-8568, Japan; Tel: +81-29-861-7218; Fax: +81-29-851-5426;
A number of ab initio [19, 20] and fragment-based [21,
E-mail: d.g.fedorov@aist.go.jp 22] linearly scaling methods have been proposed, which

1568-0266/12 $58.00+.00 © 2012 Bentham Science Publishers


2 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

compete with semi-empirical approaches [23, 24]. QM GAMESS (now renamed Firefly) [35] branched off from the
methods are indispensable for describing chemical reactions main project GAMESS-US [36]. The three branches share
and excited states, as well as for free radicals [25]. much of the source code, and the input and output file struc-
The fragment molecular orbital (FMO) method [26] is ture, but a lot of independent new development has been
one of the fragment-based methods. Impressive results can invested in each project so that they have a substantially dif-
be achieved when both linear scaling QM approaches and ferent functionality and performance. GAMESSPlus is an
supercomputing technologies are combined [27]. For exam- add-on with additional functionality for GAMESS-US, in-
ple, FMO energy calculation of the receptor-ligand system cluding solvation models, DFT functionals and QM/MM
with size of 17,767 atoms takes only ~54 minutes on Blue capability [37].
Gene/P 40,960 CPU cores at the RHF level of theory and 6- While Firefly is designed specifically for Intel-
31G* basis set [28]. With the current pace of advances in compatible CPUs, GAMESS-US and GAMESS-UK also run
both computer technology and algorithms, a wide use of QM on other types of architecture: in particular, when a UNIX
simulations for realistic biochemical systems is expected in operating system is available. Source code can be obtained
the near future. for the former two projects, and precompiled binaries are
Ligand binding is a very complex phenomenon and both provided for all three. GAMESS-UK is available for free for
enthalpic and entropic contributions need to be appropriately academic users in the UK, while GAMESS-US and Firefly
dealt with in order to estimate the free energy of binding are free for both industry and academics. WinGAMESS is a
accurately. It is important to mention that QM methods de- GAMESS-US binary precompiled for Microsoft Windows.
scribed in this paper are used exclusively for computing en- The GAMESS forum [38] can be used for discussions.
thalpy and solvation energy. Computation of configurational GAMESS can take advantage of massively parallel com-
entropy is obviously important to obtain complete energetic puters. For GAMESS-US, distributed data interface (DDI)
picture of ligand binding but it is notoriously difficult and it [39] was developed which is one sided-communication mes-
is often not practical to use quantum-chemical methods for sage passing library that works on top of either Unix sockets
this purpose. Classical molecular dynamics simulations (for or standard MPI [40]. DDI was used to parallelize various ab
example: quasiharmonic analysis with the covariance) [29] initio QM methods [41-43]. Generalized DDI (GDDI) [44]
are commonly used to estimate configurational entropy based on an efficient node grouping can be utilized with
which will not be covered in this paper. However, in medici- some methods such as FMO on large supercomputers. For
nal chemistry program, particularly in lead optimization parallelization, Firefly uses MPI in combination with the
stage, ligands of interest share a common molecular frame- thread-based Point-to-Point (P2P) message oriented inter-
work and it is often safe to assume that they the entropic face; the latter works via Ethernet (using TCP/IP sockets),
contributions cancel out. We would also like to emphasize shared memory or InfiniBand. GAMESS-UK can be paral-
that binding free energy is not the only property useful for lelized with the Global Array toolkit or MPI.
drug design. QM methods, particularly together with energy
decomposition analyses, can also provide much richer in- Here we list the QM methods which are of particular
formation on the nature of protein-ligand interactions to me- relevance to drug research. The most basic QM method is
dicinal chemists for ligand design. For example, non- Hartree-Fock (HF), which exists in the restricted (RHF) and
classical interactions such as CH- and halogen- etc can be unrestricted (UHF) varieties. The former is used for closed
readily identified, which are otherwise hard to detect and shell systems such as organic molecules and proteins, while
quantify. the latter finds it use mainly for systems with transition met-
als or radicals. Two main branches of methods improving on
In this review we focus on in silico drug design methods
HF have evolved in QM. One is DFT, which offers a similar
and applications of QM methods implemented in the quan-
computational cost to RHF, and is in general very successful
tum chemistry package called General Atomic and Molecu-
in many systems without nearly degenerate orbital spaces.
lar Structure System (GAMESS) [30, 31]. There are several
The other is given by the explicit treatment of the electron
excellent general reviews of GAMESS by its principal de-
velopers, Gordon and Schmidt [32, 33], whereas we only correlation, for which there is a well arranged ladder of
methods offering a clear way to improve the accuracy.
describe applications of GAMESS relevant to drug research.
Among these, second order Møller–Plesset perturbation the-
We cover in detail the following studies: pKa estimation,
ory (MP2) is most readily useful for large systems, although
geometry optimization, computation of ligand binding en-
more advanced methods can be also applied in some cases,
ergy, definition of quantitative structure-activity relationship
such as coupled cluster (CC) theory.
(QSAR) descriptors, and analysis of the ligand-receptor in-
teraction energies with the goal of designing new drugs with For DFT, there are many functionals whose choice is
a better affinity and specificity. We conclude with an outlook driven by experience. B3LYP [45] and PBE [46] are popu-
on a future use of QM in drug research, with the hope to lar, and among recent development we mention long-range
provide convincing arguments of the great potential held by corrected CAM-B3LYP [47], and M06 [48]. The latter per-
QM methods. forms especially well for organometallic and inorganometal-
lic chemistry and for non-covalent interactions. An ad hoc
2. QUANTUM-MECHANICAL METHODS AND way of improving DFT is to include an empirically param-
GAMESS OVERVIEW eterized dispersion correction [49], which can also be used
for RHF. For MP2, there is some choice of methods: the
The history of GAMESS is described in detail by Gordon spin-component scaled version of MP2 (SCS-MP2) [50], and
and Schmidt [33]. At some point, GAMESS-UK [34] and PC
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 3

accelerated regular MP2, based on the resolution of the iden- tions of GAMESS, one can find the study of HIV-1 by
tity (RI-MP2) [51]. Weltman et al. [57].
We focus on the methods which are particularly suitable
to calculations of large systems: the fragment molecular or- 3.2. Fragment-Based Approaches
bital (FMO) method and QM/MM, performed by interfacing The following low-scaling ab initio based methods are
GAMESS with MM programs CHARMM (Chemistry at available in GAMESS-US: effective fragment potential
HARvard Macromolecular Mechanics) [52] or TINKER (EFP) [8], elongation method (ELG) [58], divide-and-
[53]. conquer (DC) approach [59] and FMO [26]. The former is a
QM-parameterized force field and we describe it below to-
3. METHODS FOR LARGE SYSTEMS gether with other force fields. The ELG method is primarily
used for polymers, although some preliminary probing has
The choice of an optimal method depends on the type of
calculations and should pursue the balance between the accu- been performed on DNA [60, 61] and the applicability to
biochemical systems has been discussed [62]. DC methods
racy and speed. The latter is an especially relevant factor for
in GAMESS-US have so far not been applied to biochemical
drug research because of the need for providing useful in-
systems, so we do not describe them any further.
formation for ligand design to chemists during rapidly mov-
ing medicinal chemistry projects. A large ligand-receptor The FMO method [26] has been widely applied to large
system can be split into three major regions: an active site molecular systems, including a recent geometry optimization
with a ligand, the rest of the protein, and solvent. Ideally, of a 0.1 μm BN nanoring [63]. There are several reviews of
one would prefer to treat the whole system by QM, but to FMO [22, 64, 65], book chapters [66-68] and a full book on
achieve high speed and accuracy each region can be treated FMO [69], which can be used for a more detailed and con-
with different models. QM/MM methods important for drug cise study of FMO. Here we provide an overall picture with-
research are reviewed in detail below. out giving too many technical details. The relation of FMO
QM/MM is suitable for fast high-throughput calculations, to other methods has been recently discussed [22, 65] and we
do not address it here.
but for an analysis of an active site revealing ligand-receptor
interaction details, full QM methods like FMO are powerful The idea of functional groups is paramount to chemistry.
alternatives. The reason why it is desirable to perform QM This is because the functional groups, if properly defined,
calculations of the whole system can be a large ligand having retain to a large extent their physical properties in various
multiple weak van der Waals interactions with the receptor, molecular systems, under the perturbing influence of envi-
multiple successive chemical reactions in the active site, or ronment (e.g., other functional groups). This is exactly the
an excited state involving delocalized CT. basic principle behind fragment-based methods such as
FMO. By taking advantage of chemical knowledge, one can
3.1. Full ab initio define groups of atoms, called fragments, which are de-
scribed by QM, and an interaction between them. Clearly,
In addition to a high computational cost of traditional ab the success of such an approach relies on the efficiency of
initio methods, the memory requirements are also high, es- treating the interactions.
pecially for MP2 and CC. Even in the most economic RHF
and DFT, the memory requirement scales quadratically with In FMO, the electrostatic interaction is accounted for by
the system size, which eventually becomes a problem. Usu- including the polarizing electrostatic field in all fragment
ally, there are several matrices one has to store. Assuming calculations. The field corresponds to the whole system, and
for simplicity that one stores the Fock, density, overlap and it is computed using the electron densities and nuclei of all
MO matrices, this gives 48N2 bytes, where N is the num- fragments (called monomers). The other key step is to calcu-
ber of basis functions. For N=10000 (approximately 1000 late pairs of fragments (dimers) using QM, in order to de-
atoms with 6-31G*), this is 3.2 GB. Unless the matrices are scribe non-electrostatic interactions, such as exchange-
distributed among computer nodes, this amount of memory repulsion and CT. Alternatively, there is also effective FMO
is required per CPU core, plus some memory is taken by OS (EFMO) [70], which differs from regular FMO in the de-
and the program itself. Even for a modern workstation 4 GB scription of the electrostatics. EFMO can be used in many
per CPU core is a significant amount, and large supercom- ways similar to FMO, for example, to do QM calculations of
puters such as Blue Gene typically have very moderate proteins [71].
amount of memory per core. Treatment of covalent bond detachment during fragmen-
Thus, Alexeev et al. [42] developed a self-consistent tation is a technical issue well described elsewhere [65, 69].
field (SCF) algorithm for RHF in GAMESS-US, in which Here we only mention that the electron density distribution
large matrices are distributed between nodes. On the other of a detached bond is assigned to a single fragment, i.e.,
hand, to overcome steep computational cost of QM, one can bonds are detached heterolytically and without the addition
use the fast multipole method (FMM) [54] to accelerate two- of hydrogen caps. The detached bonds are saturated by the
electron integrals in large systems, which was implemented embedding potential. The methodology ensures that for
[55] in GAMESS-US. Ishimura et al. developed an efficient closed shell systems the fragments are also closed shell, and
OpenMP/MPI parallelization [56] capable of using many neutral unless some charged functional group is present.
nodes in ab initio calculations. Despite this method devel- For practical applications, an important question is, how
opment, systems of biological size are still hard to treat with to define fragments? There is a rather simple answer for
fully ab initio methods. Among very few published applica- FMO, and the main principle is to minimize the CT between
4 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

fragments. This is because CT is accounted for in FMO with


dimer calculation, i.e., it is limited to two-body contributions
only. Charge transfers between pairs of fragments are cou- (1)
pled, i.e., there are many-body CT effects. The same applies First, the electronic state of each fragment is calculated
to exchange-repulsion (a quantum effect originating from the self-consistently with the polarizing embedding potential
Pauli exclusion principle) and the dispersion. All of these describing the rest of the system. This gives the energy of
effects are typically localized so that one can meaningfully fragments EI. Next, the electronic state of fragment pairs is
define fragments. When CT is not localized, then one cannot computed once in the embedding potential, producing EIJ.
fragment the system, e.g., conjugated systems such as aro- These energies are assembled in Eq. 1 above, giving the total
matic ring systems or donor-acceptor metal complexes. For energy of the system E. By taking an analytic derivative [73,
the latter type, assigning a metal cation as a separate frag- 74], one can easily obtain the gradient, which can be used for
ment will result in every substantial CT between the metal geometry optimizations [75-77]. We note that the total gra-
and ligands, thus leading to a poor accuracy. Salt bridges dient E is with respect to the atoms in the whole system,
between positively and negatively charged amino acid resi- and once the vector is constructed, it can be used in any
dues are also prone to a large CT, and one can combine them standard optimizer. Alternatively, one can define a frozen
into a single fragment for a better accuracy. domain (FD) in the FMO/FD method see Fig. (1D) [77] tak-
Biochemical systems are composed of a few dozens of ing advantage of the fragmentation, as described in more
standard units, enabling an easy automatic fragmentation. detail below.
The caveat here is that for polypeptides, there is a consider- The computational cost of FMO is determined by the
able CT across a peptide bond, so that one cannot fragment number of fragments N, and the number of dimers N2. By
it. Instead, polypeptides are fragmented at C, and residues taking advantage of the spatial separation [78] between
in FMO become residue fragments shifted by one carboxyl fragments, all dimers are divided into two groups. For the
group relative to the conventional residues. To distinguish first group with a short separation, one has to perform QM
between the conventional residues and residue fragments, we calculations. The number of such dimers scales nearly line-
often insert a dash in the latter, e.g., Trp-6. Polysaccharides arly with the system size, because for each fragment there is
are fragmented into sugar units, and nucleic acids such as a small number of other close fragments, and this number is
DNA or RNA into bases. Such automatic fragmentation can independent of N. The number of remotely separated dimers
be readily performed using Facio [72]. scales quadratically, however, their calculation does not re-
Ligands leave some freedom to the user. There are two quire QM calculations because the interaction is essentially
reasons to divide a large ligand into several fragments. If a electrostatic in nature. Thus the overall scaling of FMO is
ligand is large, it takes relatively long to compute, especially usually nearly linear, and there is an ongoing work to im-
for correlated methods such as MP2. Also, ligands often prove it [79, 80].
have some subunits and it desirable to analyze the properties Many common methods have been made available for
of these subunits such as the interaction energy between a use in FMO, such as MP2 [81] or DFT [82]. Free radicals
piece of the ligand and a residue fragment in a protein. This can be described with both restricted [83, 84] and unre-
naturally shows the connection between FMO and fragment- stricted [85] open shell methods. For instance, transition
based drug design (FBDD). The small molecules (<300 metal containing enzymes or complexes can be created with
dalton) used in FBDD can be treated as FMO fragments dur- the unrestricted FMO method. For a detailed description of
ing hit optimization and, potentially, structure-based drug available methods, refer to the recent review [65]. As pointed
design (SBDD). When fragmenting ligands, one should be out above, FMO does not give the same total energy as a
careful about CT, and ligands are often difficult to divide regular ab initio QM calculation. This is because of some
because they can be composed of extended aromatic systems many-body effects neglected in FMO. The above equation
with delocalized electrons. Technically, the division of corresponds to the two-body FMO (FMO2); higher-order
ligands is usually a manual process (unless ligands are poly- FMO3 [86] and FMO4 [87] greatly improve the accuracy at
peptides or other standard systems). Programs such as Facio the additional computational cost of computing trimers and
[72] greatly facilitate it requiring only clicking on the bonds tetramers, respectively. The accuracy of FMO depends on
which should be detached in forming fragments. the appropriateness of fragmentation and also on the basis
When in doubt, one can always do a back check. Because set (large basis sets especially with diffuse functions have
this is useful, especially for experiments with novel systems, lower accuracy, caused by large many-body effects). A bet-
we describe it briefly. After an FMO calculation is per- ter accuracy can be obtained with larger fragments, for ex-
formed, one gets the properties such as the CT between ample, a division into two residues per fragment is often
fragments. By looking at the values of CT, one can form a used for accurate energetics. In many systems containing
fairly good idea if the fragmentation was successful. Typical several hundreds of atoms the error of the FMO total energy
CT amounts for a hydrogen bond are 0.02-0.05 electrons or E was shown to be on the order of several kcal/mol com-
so. Any CT on this order should be taken as appropriate in pared to full ab initio, but it can be worse if there is a large
general. Larger values should be given some thought, espe- CT. The accuracy of the relative energies, e.g. a protein-
cially if they exceed 0.1 electron. ligand interaction energy, should be expected to be much
better than the accuracy of the absolute energies.
Next, we give the basic FMO equation and briefly de-
scribe how an FMO calculation proceeds. There are two major advantages of fragmentation. Firstly,
fragmentation has high computation efficiency, because only
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 5

small subsystems (fragments and their pairs) are computed, A general potential EFP2 [94] was developed with an
which can also be efficiently parallelized. Inexpensive PC automatic generation of its parameters, which is in principle
clusters consisting of several nodes connected by Gigabit applicable to any system. Both EFP1 and EFP2 include the
Ethernet can nowadays have a hundred CPU cores or more, effects of the polarization and CT, and EFP2 also has an ex-
providing a powerful tool for applications of FMO to drug plicit dispersion [97]. EFP2 is fully developed for EFP-EFP
design. On the high end, FMO can be used on supercomput- interactions, and a QM-EFP2 interface is also available. The
ers [27] taking advantage of many thousands of CPU cores. performance of EFP2 has been demonstrated on a variety of
The second advantage is the fragment properties computed molecular systems [98, 99]. EFP1 was also applied to pro-
along with the total energy E in the FMO method. Some of teins [100], using frozen localized orbitals on fragment
these properties are automatically generated in any calcula- boundaries. EFP1 was interfaced with FMO [101] and can be
tion, and some are optional requiring additional computa- used to describe explicit solvent in FMO.
tions. The fragment properties in FMO are: multipole mo-
ments of fragments (dipole, quadrupole etc), fragment po- 3.4. QM/MM Methods
larization energies and molecular orbitals of fragments. The
properties of fragment pairs are the pair interaction energies There are few conceptual choices to accurately compute
and the CT between them. In solution, there are also the cor- chemical reactions in the active site of a protein. One is to
responding solute-solvent properties. The total properties apply ab initio QM methods which would limit the size be-
include the energy, its gradient, atomic charges and molecu- cause of the computational time and memory limitations. A
lar orbitals. more attractive option is a compromise: the active site is
treated accurately with QM while the rest of the system is
Now we describe the pair interaction energies at some treated with MM. The QM region is usually chosen to be
length. As can be inferred from eq. 1, the difference EIJ-EI-EJ large enough [102] to include all important electronic struc-
is the pair interaction energy between fragments I and J. It ture effects. The MM region introduces structural constraints
also includes the embedding potential (because each of the and environmental effects (electrostatic and van der Waals
fragment or dimer energies is computed with it), and thus interactions) acting upon the QM region. There are some
contains explicit many-body effects, complicating the analy- disadvantages of QM/MM: a number of successive chemical
sis. Therefore, one usually uses a different form of the pair reactions in an active site may force a very large QM region
interaction energy based on the internal energies of frag- or a multiple redefinition of a smaller one. Also, artifacts in
ments and their dimers (i.e., with the embedding potential the QM/MM interface may lower the accuracy of this ap-
separated). In this case one obtains the total values, which proach. In addition, there is the complexity of setting up cal-
can be decomposed further into components using the pair culations. The coupling between the QM and MM regions is
interaction energy (PIE) decomposition analysis (PIEDA) the central issue of any QM/MM method defining its accu-
routines implemented in FMO [88-91]. racy and general usefulness.
There are a large number of implementations of
(2)
QM/MM, but we cover only those relevant to GAMESS in
The components of the total values (INT) are: the elec- chronological order. Detailed QM/MM reviews can be found
trostatics (ES), exchange-repulsion (EX), CT and higher elsewhere [103-105]. Firstly we mention the implementation
order terms (CT+mix), dispersion (DI) and solvent screening by Field et al. [106] of the QM/MM interface between
(SOLV). The latter component [91] is defined for the po- CHARMM and QUANTUM codes, which was later ex-
larizable continuum model (PCM). The polarization is an tended to CHARMM/GAMESS interface by Lee et al. [107].
electrostatic effect, and it can also be computed in PIEDA The partitioned system is described by the effective Hamil-
with the definition of a reference non-interacting state, and tonian:
monomer energies EI, which are affected by the polarization.
As described below, pair interaction energies between ligand (3)
and protein residues provide useful descriptors in QSAR.
where and are the Hamiltonians of the QM and
3.3. Molecular Mechanics MM regions, respectively. represents the electrostatic
Standard force fields, such as Amber [3, 4], CHARMM and van der Waals interactions between the QM and MM
[52], OPLS [92], and MMFF [93], can be used in GAMESS atoms, whereas describes periodic or stochastic
via an interface to CHARMM or Tinker. In addition to that, boundary effects in the system. The overall setup is shown
there is also a more advanced ab initio derived force field, on Fig. (1A). To treat a covalent bond between QM and MM
EFP. There are many excellent reviews of EFP elsewhere atoms, link atoms are utilized, which are treated as regular
[22, 94], and here we only briefly mention its main points. QM atoms interacting with the atom charges in the MM re-
EFP is a rigid model, i.e., its internal geometry always re- gion. In CHARMM, link atoms have no charge, no van der
mains frozen, although fragments can move relative to each Waals parameters and they do not appear in the MM energy
other. The original EFP (now called EFP1) was parameter-
terms. The described approach was extensively tested on
ized for water by fitting its terms to ab initio calculations of
small systems and it was found that depending on the parti-
water dimer as a function of distance [8]. EFP1 was success-
tioning to QM and MM regions one can well reproduce full
fully applied to a wide array of problems [95, 96] modeling
solvent effects on chemical reactions in ground and elec- QM geometries [108]. Although this is a well-tested method
tronically excited states. to treat QM/MM interface, the alternative frozen orbital
6 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

method [109] is considered more accurate [103]. Cui and QMMM program developed by Truhlar and coworkers
Karplus [110] developed analytical second energy deriva- [117], which supports GAMESS-US.
tives for QM/MM, which enables an evaluation of vibra-
tional frequencies and infrared intensities in large systems. 3.5. Solvation Models
Solvation effects can play a crucial role in the structural
and dynamical properties of a ligand interacting with the
receptor. The solvation effects, broadly divided into non-
polar and electrostatic contributions, can be described by
explicit or implicit solvation models or their combination
called the supermolecule approach [118, 119]. In QM meth-
ods, the implicit solvation models usually rely on solving the
Poisson equation [120] numerically. The implicit solvation
models implemented in GAMESS-US are the Onsager cavity
model or self-consistent reaction field (SCRF) [121], con-
ductor-like screening model (COSMO) [122], PCM [120],
surface and simulation of volume polarization for electrostat-
ics (SS(V)PE) [123], and solvation model 6 with temperature
dependence (SM6T) [124]. GAMESS-UK supports SCRF,
PCM, and direct reaction field (DRF) [125]. In Firefly, one
can use SCRF or PCM. On the other hand, explicit water
molecules can be included in the QM region, or treated by
MM (EFP).
The implicit solvation models are computationally
cheaper compared to explicit description of water, but they
sometimes have poor accuracy, especially if water molecules
in the first solvation shell form strong hydrogen bonds with
the solute. Water molecules often coordinate a ligand and are
vital for a proper description of the structure and energetics.
Fig. (1). Different composite methods. (A) QM/MM in GAMESS/ In the supermolecule approach the first solvation shell is
CHARMM. (B) QM/MM/PCM in GAMESS/CHARMM. (C) described explicitly, however, a statistical sampling of water
QM/EFP/PCM in GAMESS. (D) FMO/FD implemented in configurations should be performed, for example, based on
GAMESS, where AD and FD are active and frozen domains, re- molecular dynamics or Monte Carlo methods. Water mole-
spectively. cules can be treated with MM instead of QM, for example, in
QM/EFP.
For an efficient geometry optimization Maseras and A popular implicit model for interfacing with QM/MM
Morokuma proposed a variation of QM/MM called the inte- methods is PCM, where a cavity surrounding the solute is
grated molecular orbital molecular mechanics (IMOMM) formed by combining spheres centered on each atom of the
[111] which was later implemented in GAMESS-US via an solute. The cavity surface is divided into many tesserae, and
interface with Tinker by Shoemaker et al. [112] in their sur- each tessera has its own surface charge determined by the
face IMOMM (SIMOMM) method. IMOMM differs from electric field on it. The charges on the tesserae are solved
QM/MM in neglecting the electronic embedding (there is no self-consistently with the electronic state of the solute. Con-
polarization of the QM region by MM). sequently, those charges make direct contributions to the
EFP is interfaced with most of QM methods imple- energy and the gradient of the solute. In PCM the free energy
mented in GAMESS-US. As in all QM/MM methods cova- includes the following solvent-related term :
lent bond division represents a challenge, which in EFP was
addressed by utilizing frozen localized molecular orbitals (4)
(LMOs) [100]. In QM/EFP one introduces the buffer region
with LMOs frozen during SCF, see Fig. (1C). Computations where the electrostatic (es), repulsion (rep) and dispersion
of the proton affinity with QM/EFP for a number of small (disp) solute-solvent interactions are added to the cavitation
organic systems demonstrated that this method consistently (cav) energy. The PCM can be combined with most QM
reproduced full ab initio values within 0.4 kcal/mol [100]. methods. The main problem of PCM is the parameterization
Nemukhin and coworkers proposed to use flexible fragments of atomic radii which defines the cavity shape. There are a
in QM/EFP to treat biochemical systems [113, 114]. Alterna- number of sets of radii, such as van der Waals radii or
tively to QM/EFP, one can do QM/MM calculations in United Atom for Hartree-Fock (UAHF) [126]. Depending on
GAMESS with polarizable force fields using QuanPol [115]. the choice of the radii, the results can significantly differ.

There are few QM/MM platforms which support multiple Cui et al. interfaced PCM with QM/MM implemented in
QM and MM packages. In particular, ChemShell developed GAMESS/CHARMM [127], see Fig. (1B). It can be consid-
by Sherwood et al. [116] currently supports many QM pro- ered a variant of ONIOM-PCM [128]. One of the ways to
grams including GAMESS-UK. Another platform is the apply QM/MM/PCM is to treat the ligand with QM, and the
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 7

rest of the receptor and the water molecules in the first solva- calculations, automatic protonation and other capabilities. A
tion shell by MM (a supermolecule approach), and describe set of tools for FMO (FMOtools package) [138] is developed
continuum by PCM. QM/MM/PCM was applied to study by Alexeev specifically for FMO implemented in GAMESS-
small model systems. In particular, the relative stability of US. The package has a Web interface and consists of four
the neutral and zwitterionic forms of glycine in solution was modules (FMOcheck, FMOsolv, FMOgen and FMOres)
correctly reproduced [127]. which allow checking PDB structure for consistency errors,
fragmenting proteins and ligands, visualizing fragments in-
Bandyopadhyay et al. interfaced EFP with PCM [129] in
teractively in Jmol, solvating the system, generating
a similar way as Cui interfaced PCM with QM/MM. Later,
GAMESS input files, running short GAMESS-US simula-
Li et al. implemented a more computationally efficient and
tion to test input file syntax, analyzing and visualizing
parallelized version of the EFP/PCM solver [130]. In the
latter case, EFP-induced dipoles and PCM-induced charges GAMESS-US output (including the pair interaction ener-
gies). FMOtools is the only package which supports input
are iterated to self-consistency during convergence of the
generation and analysis for the FMO/FD method. FragIt is
electronic state. The analytical gradients were developed also
developed for a smart fragmentation of large molecular sys-
[119] enabling an efficient geometry optimization.
tems based on chemical patterns [139, 140].
QM/EFP/PCM model shown in Fig. (1C) was successfully
applied to study stability of glycine in solution [129] and pKa There are many graphical programs which support multi-
prediction [131]. ple GAMESS packages and other QM programs. One of
such programs is Facio [72] which is freely distributed as a
PCM was interfaced with FMO by Fedorov et al. [132].
precompiled binary. Facio supports GAMESS-US and Fire-
Later the analytical gradient was implemented [133]. In
fly and can do molecular modeling, build input files, and
FMO/PCM the solvent is treated as in other QM/PCM ap-
visualize results. It has a very advanced GAMESS/FMO
proaches: there is a solvent cavity surrounding the whole
solute, and in each fragment calculation one has to add the interface, which can automatically fragment polypeptides,
saccharides, nucleotides or any combination thereof, possi-
electrostatic field of the point charges on the cavity. These
bly in the presence of other standalone molecules (ligand or
charges are induced by the electronic state of the solute, rep-
explicit solvent). It can also manually fragment ligands or
resented by the electrostatic potential. In FMO, it is com-
other systems, by clicking bonds to be detached. The results
puted from monomer and dimer contributions. Several levels
of FMO calculations can be visualized, in the form of inter-
of FMO/PCM were developed for the many-body expansion
of this potential. The cheapest level of PCM[1] is based on active maps and graphs of pair interactions, with a clear link
to the structure. In difficult cases, for instance, if metal
only monomer contributions to the solute potential inducing
cations or metal-coordinated fragments are present, the user
solvent charges. In PCM[1(2)], dimer corrections are also
may need to intervene and manually change the automatic
added in the perturbative fashion to PCM[1]. The former
fragmentation to prevent excessive CT.
method is often used for an accurate evaluation of the ener-
getics while the latter is mainly used in geometry optimiza- There are other universal graphical programs such as
tions. Recently, an improved PCM1 model for FMO was Winmostar [141, 142], Avogadro [143], Molden [144],
developed by Nagata et al. [134], which is both accurate and ChemCraft [145], GDIS [146], MaSK [147], and Gabedit
efficient, being a modification of PCM[1]. [148], which offer some support of GAMESS programs.
CCP1 GUI [149] is an advanced graphical interface for
In PIEDA/PCM, the physical picture of the solvent
GAMESS-UK, specifically developed to support many of its
screening in PCM was elucidated [91]: the screening arises
because of the charges induced on the solvent, which com- features. PC-GUI [150] and Ascalaph Quantum [151] are
graphical interfaces for Firefly. Furthermore, there is an
pensate the direct Coulomb interaction between charges in
FMO interface extension to the drug discovery software
the solute. In other words, those induced charges are directly
package MOE that is available from the SVL Exchange
related to the dielectric constant. The local screening can
[152]. This GUI allows for the preparation of FMO input
take values, very different from those determined by the bulk
files for GAMESS and for the visualization of the output
macroscopic constant, thus the actual interaction picture in
the solute has many local features determined by its local from FMO, such as PIEDA.
electrostatic field, and often unexpected results are seen be- Using web interfaces for visualizing molecules, generat-
cause of the of the many-body electrostatic effects. Details of ing input files, and analyzing results is an exciting new de-
these interactions can be readily studied with FMO/PCM, velopment owing to the rapid development of web technol-
based on fragments and their induced charges. ogy in recent years. An appealing advantage for users is that
there is no need to install and update software. A number of
4. SOFTWARE TOOLS graphical viewers for 3D chemical structures were written in
Java which allows running program as a web browser applet
There are a number of graphical programs to generate and standalone Java application on all modern operating sys-
input files and visualize GAMESS output. MacMolPlt [135] tems. One of such programs is Jmol [153] which was used in
was developed exclusively as a graphical user interface for the FMOtools interface [138]. WebMO [154] also supports
GAMESS-US. It allows preparing input files, displaying and GAMESS.
animating molecules, normal modes of vibrations, reaction
paths, as well as plotting electron densities and molecular
5. MODELING MOLECULAR SYSTEMS
electrostatic potentials. Precompiled binaries can be down-
loaded for free [136]. FMOutil [137] offers a simple inter- The initial preparation of the geometry structure and pro-
face to converting PDB files into GAMESS input for FMO tonation is the first crucial step in drug design. The accuracy
8 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

of the predicted ligand binding affinity is directly related to ionizable amino acid is directly transformed from the proto-
the quality of the geometric structure of the ligand-receptor nated form into the deprotonated form with FEP. Thus, FEP
complex. The correct protonation of the ionizable amino computes directly the difference of the solvent environment
acids plays an important role in the prediction of protein sta- interaction of the amino acid from the protonated form into
bility, binding ability, and catalytic activity. Since the initial the deprotonated form. As a result this approach is more ro-
preparation can affect the mechanism of a chemical reaction bust because the variations in the electronic structure and
or the binding of a ligand to the receptor, accurate methods nuclear geometry during the protonation of the amino acid
such as FMO and EFP can be helpful in a reliable initial are computed accurately [167]. But there are disadvantages:
preparation. relatively high computational cost, the accuracy dependence
on a number of parameters, and having to treat multiple titra-
5.1. pKa Prediction tion sites. But for certain systems such as proton pumps and
redox reactions [168] this is a promising approach.
The early approaches [155] to compute pKa in proteins
considered only the interactions between the charged groups The described QM/MM-FEP method was applied by Cui
represented as point charges in dielectric continuum, and et al. to compute pKa for CH3 CH2SH in water [166]. The
solved the electrostatic Poisson-Boltzmann equation. These experimentally determined pKa is equal to 10.6 [169]. By
approaches failed to take into account other important inter- using the self-consistent charge-density functional tight bind-
actions, notably the hydrogen bonds which affect pKa of ing (SCC-DFTB) model [170] and the water molecules de-
ionizable amino acids. Besides that, the description of the scribed by TIP3P model [171] the computed pKa was equal
interactions by point charges ignoring the electronic struc- to 7.7. In principle, SCC-DFTB can be replaced by a QM
ture can produce a significant error. One solution is to treat method in GAMESS via the Q-Chem module in CHARMM,
the short-range interactions with QM and the long-range which may improve the results.
interactions with MM in the QM/MM approach. MM can use
a classical force field or the more sophisticated EFP. As a 5.2. Geometry Optimization
way to improve the speed of calculations, EFP fragments far There are several options for geometry optimization of
from QM region can be replaced by force fields as shown in
large systems in GAMESS: IMOMM, QM/MM and FMO. It
Fig. (1C).
is important to consider the level of accuracy, the cost of
Such approach was used to predict the pKa value of calculations, the simplicity of computational setup and the
Lys55 in protein OMTKY3[156], where the following way reliability of solvent models. In the following, we focus on
was used to compute pKa to eliminate the systematic error in the use of FMO.
the free energy. For the reaction,
FMO can be applied to optimizations of biochemical
Lys55H++ CH3NH2  Lys55 + CH3NH3+ (5) systems either in its usual form or with the frozen domain
(FMO/FD); in addition, there is an ongoing work [75] to
the following relation was used,
develop FMO/MM. For the regular FMO, where one treats
pKa (Lys55H+) = 10.6 + ([G(Lys55) - G(Lys55H+)] - the whole system with FMO and optimizes the system using
[G(CH3NH2) - G(CH3NH3+)])/1.36 (6) the total energy gradient, the computational cost is aggre-
where 10.6 is the experimentally determined pKa of methy- gated by the fact that most biochemical systems are very
lamine (instead of methylamine any chemical compound can flexible having flat energy surface, and require hundreds or
be used, which donates or accepts a proton with an accurate thousands of single point calculations if all degrees of free-
experimentally known pKa). The free energy G of each com- dom are optimized. We briefly mention without further dis-
pound is a sum of the ground state electronic energy and the cussion the existence of the problem of finding the global
solvation energy. Only Lys55 and Tyr20 side chains in OM- minimum, and the need to sample other local minima, both
TKY3 were treated by QM and the rest of the system was of which are of major concern when explicit models of sol-
treated with an EFP. The computed pKa for Lys55 was 11.4 vation are used.
which agrees well with experimentally determined pKa of The Trp-cage miniprotein (304 atoms) was optimized
11.1 [157]. More details can be found elsewhere [158] in- with FMO using MM [75] or PCM [133, 134] to treat the
cluding further applications [131, 159-162]. solvent. When the dispersion is included, a good agreement
The work is this area inspired Jensen and coworkers to to experiment is obtained: RMSD to NMR experiment was
develop a simpler and faster pKa empirical prediction method 0.414 Å [134]. In addition, the helical oligosaccharide hepa-
[163] implemented in the program PROPKA [164]. The rin was also optimized in solution [172] (RMSD to NMR
root-mean-square deviation (RMSD) of predicted pKa from experiment: 0.586 Å) and fluorinated prostaglandin receptor
experimental values was found to be less than 1 [163]. was optimized in gas phase [173].
PROPKA can also predict pKa for protein-ligand complexes FMO/FD was developed to efficiently optimize an active
[165]. Once pKa of all ionizable sites are known, it is easy to site in protein-ligand complexes. In this method, one takes
predict the protonation states for a given pH. advantage of the fact that only atoms in the active site can
An alternative approach to computing pKa of amino acids move during geometry optimization. Therefore, it is not nec-
was proposed by Cui et al. [166], who combined QM/MM essary to recompute the whole protein at each step. The
with the free energy perturbation (FEP) method (QM/MM- whole system is divided into the polarizable buffer Fig. (1D)
FEP). The thermodynamic cycle is different from that used and frozen domain (the rest). The polarizable buffer includes
in QM/EFP. One of the most important differences is that the the active domain (e.g., the ligand and the binding pocket of
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 9

a protein) and it also includes some part of the protein sur- ple ligands. To understand the interaction mechanism of dif-
rounding the active domain. This is done because the elec- ferent ligands PIEDA can be a valuable tool. FMO/FD can
tronic state of the polarizable buffer should adjust to the be used in other applications. For example, if an X-ray struc-
changes in the coordinates of the active domain. ture of a receptor-ligand complex is available at low resolu-
FMO/FD does calculate the whole protein once for the tion then FMO/FD can be used to optimize geometry of the
ligand.
initial geometry at the level of FMO1, i.e., by self-
consistently polarizing all fragments in their total electro-
static field. For all consequent geometries, only the elec- 6. DRUG DESIGN RELATED STUDIES
tronic state of the polarizable buffer is recomputed, in the While force field applications to drug research in the
field of the fixed electronic state of the frozen domain. There form of MM [2] and QM/MM [11] have been numerous and
is a tracking of the movement of the polarizable buffer, so fruitful, a number of researchers pointed out the advantages
that if necessary, one can recalculate the electronic state of of QM based approaches [185-188] in studying biochemical
the frozen domain, or redefine the domains. systems or demonstrated the efficiency of quantum refine-
In addition to these direct cost reductions, one can further ment [189]. Fujimura and Sasabuchi argued that many com-
improve the description by using a different basis sets and/or monly used force fields such as AMBER may not be reliable
wave functions for the frozen domain and polarizable buffer. for treating halogen atoms [173]. Ozawa et al. [190] dis-
For the former, a very cheap level of RHF/STO-3G can be cussed the role of CH- interactions in ligand recognition,
used, while for the latter a better level of DFT-D/6-31G* can based on FMO calculations. Chuman and coworkers [191-
be recommended (DFT-D is DFT with empiric dispersion). 193] used FMO for QSAR-related studies.
This is the level used in the demonstrative optimization of Ishida et al. [194] discussed the collective catalytic de-
prostaglandin H(2) synthase-1 in complex with the reversible vice in the chorismate mutase catalyzed reaction, using
competitive inhibitor ibuprofen (PDB: 1EQG), containing QM/MM and FMO calculations. DNA and estrogen receptor
19471 atoms, which was accomplished in 32 hours on 6 interaction was analyzed with FMO by Watanabe et al.
desktop computers using quantum mechanics for all atoms [195]. Fluorescent proteins find various use in biology and
[77]. Mochizuki and coworkers [196] showed the efficiency of
FMO/FD calculation requires some user input to define FMO in describing the excited states of red fluorescent pro-
the domains. The pilot calculations [77] provided guidelines teins. Zwier and coworkers [197] demonstrated how PIEDA
based on the optimizations of the Trp-cage miniprotein com- can be used in analyzing amide stacking in -polypeptides.
plexes with a neutral and charged ligand. The latter type re-
quires larger domains for the same level of accuracy, be- 6.1. Use of FMO in Binding Energy Calculations, Pro-
cause the polarization is longer-ranged. FMOtools [138] tein-Ligand Interactions, and QSAR
have a convenient interface to set up FMO/FD calculations, FMO is a fully QM alternative to QM/MM calculations.
including a script processing for an automatic pipeline Here, we briefly introduce applications of FMO to protein-
stream of ligands. ligand complexes. Two general types of studies can be
In comparison to QM/MM, FMO/FD describes the po- named. (a) Analysis of systems, for which the binding is
larization of the outer (frozen) buffer, and does not need pa- known experimentally, and it is important to analyze how the
rameters for the ligand, whereas in case of MM the problem ligand is recognized by the protein. This often gives hints to
of having to provide force field parameters for ligands can be possible mutations in the protein or to choose a different
very complicated. It should be noted that QM methods in ligand in order to increase the binding. In addition, a good
general are quite unforgiving for poor initial structures, agreement to experiment can be used to justify the computa-
which can happen when raw X-ray or NMR structures are tional model. (b) Predicting the binding affinity of ligands,
taken, resulting in divergence of SCF. Therefore, it is neces- for which no experimental data are available. This can be
sary to prepare an initial structure of the receptor (protein), used for screening ligands, and discovering new potent
which is relatively easy to do with force fields such as drugs. FMO can be utilized for both of these study types.
CHARMM or AMBER. X-ray structures often have an in- Nemoto et al. [198] used multilayer FMO in studying
sufficient resolution with ambiguous binding poses, and ligands bound to pheromone-binding protein. Nakanishi et
therefore a reliable optimization of these structures is neces- al. [199] elucidated the molecular recognition mechanism of
sary. the FK506 binding protein by combining gas phase FMO
Next, receptor structure can be docked with ligand using and solvation energies from MM models. Sawada et al. [200,
one of the docking programs like Autodock [174], Autodock 201] complemented FMO/PCM results for a single structure
Vina [175], Dock [176], GOLD [177, 178], FlexX [179], with MM-derived entropic contributions in their analysis of
FRED [180], Surflex [181], GLIDE [182, 183], and ICM the binding between sialosides and influenza virus hemag-
[184] to generate binding poses. There is a question of reli- glutinin. These studies clearly demonstrate that gas phase
ability of the ligand docking by these protocols. It is con- calculations lead to a drastic overestimation of the binding
ceivable in future to develop QM-based docking, by either because of the desolvation penalty [202], i.e., the energy cost
rescoring the poses using FMO energy function or using to partially desolvate the protein and ligand surfaces during
faster FMO methods such as FMO/FD to generate new their complex formation. Sawada et al. [200, 201] also dem-
poses. The most promising poses can be optimized at a onstrated that FMO can be applied to large size systems at
higher QM level. This procedure can be repeated for multi- the solvated level with electron correlation (FMO-
10 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

MP2/PCM/6-31G*): the protein-ligand complex treated at • assigning the optimal fragment binding pose present in
this level contained 24060 atoms. crystal structures for fragment-based drug discovery,
A by-product of the total binding energies with FMO is • decomposing a ligand into fragments to study the struc-
the pair interaction energy analysis, which gives the inter- tural contributions to binding free energy,
fragment interaction energies, for example, between amino
• assigning the correct ligand binding pose amongst mul-
acid residue fragments and the ligand. Here, some caution in
tiple similarly scored poses from virtual screening,
comparison to other methods such as MM should be exer-
cised, because fragment residues differ from conventional • predicting the correct binding pose in poorly resolved
residues by one carboxyl group, due to the electron delocali- X-ray crystal structure electron densities.
zation in the peptide bonds, as mentioned above. Using this A challenge for computational chemistry support of in-
analysis, important insight on the ligand recognition can be dustrial rational drug design efforts is to find the appropriate
obtained. The total pair interaction energies can be further balance between the level of complexity and the computa-
decomposed into physical components in PIEDA, as de- tional overhead required to understand drug-receptor interac-
scribed above. tions within the timescale necessary to drive medicinal
Ohno et al. [203] took advantage of PIEDA in their de- chemistry. The highly parallelized FMO calculations using
tailed study of the binding of biphosphates to farnesyl pyro- the 6-31G* basis set with the second order Møller-Plesset
phosphate synthase. Merz et al. [204, 205] demonstrated that (MP2) perturbation method provide a practical solution to
FMO is efficient in predicting the native conformations of achieve the required accuracy and speed. This moderate level
proteins, and an even better prediction can be obtained with of theory in combination with the FMO methodology allows
parametrized corrections of the basis set superposition error. for tens of calculations to be run per day on a relatively small
Prime et al. [206] used PIEDA as a tool in their discovery cluster (32 CPU cores and 3G of RAM per core) sufficient to
and SAR of potent and selective covalent inhibitors of trans- achieve the required accuracy and effective timescale to im-
glutaminase 2 for Huntington’s disease. pact medicinal chemistry projects related to the analysis of
protein-ligand interactions, although for massive screening a
In classical QSAR studies the descriptors are often com-
much faster time scale is needed. The advantage of MP2
puted using empirical formulas based on the structure and
includes the ability to observe the dispersion type interac-
the connectivity of atoms in the molecule. More accurate tions of aromatic and alkyl chains often seen in protein-
electronic structure descriptors can be derived from QM cal-
ligand interactions as well as halogen- interactions which
culations using FMO, for example, the highest occupied and
are difficult to capture with MM force fields alone.
the lowest unoccupied orbital energies. But FMO can also be
used to prove new kinds of descriptors to improve the accu- SARs can be analyzed and the design of new molecules
racy of QSAR. For example, Chuman and coworkers [192] can be driven using the FMO methodology. This process is
suggested using CT and pair interaction energies between the optimal when the initial start-points for the calculations are
ligand and amino acids in the protein, as descriptors in derived from X-ray crystal structures. The alternative
QSAR. Zhang et al. [207, 208] used FMO in the three- method to obtain starting points for ab inito calculations
dimensional quantitative structure-activity relationship (3D- would be to take an average structure over a number of mo-
QSAR) combined with the comparative molecular field lecular dynamics snapshots. Therefore an assumption is
analysis (CoMFA) to study receptor-ligand binding models. made that the conformation of the X-ray crystal structure is
likely to contribute significantly to the Boltzmann-averaged
6.2. FMO Applications in Drug Discovery potentials for the free energy estimation. And that a single
point calculation is more likely to be a good representation
Accurate estimation of the affinity of a small ligand to a of the system than one from which the phase space is poorly
macromolecule is extremely challenging. Many methods sampled. This method has been used previously to generate
have been developed to this end, ranging from rigorous but poses for FMO calculations [209]. Other studies have used
time-consuming FEP and thermodynamic integration (TI) MM/MD simulations ascertain an optimal system conforma-
simulations to fast empirical scoring functions. However, tion before further analysis using FMO [210].
even FEP and TI methods are only accurate within the limit
of the parameterization of the force field used. In recent For example, FMO can then be used in a combinatorial
years, it has been reported that intermolecular forces that can fashion to study the expansion of fragment- and lead-like
only be properly treated by quantum mechanical methods are hits, to explore fragment merging, functional group replace-
playing important roles in bimolecular interactions. For this ments and scaffold hopping via multiple calculations based
reason, in many occasions in drug discovery, it is highly de- around a single crystal structure. Fragment linking strategies
sirable to study the ligand/protein complex using quantum can also be realized in circumstances where the receptor pre-
mechanical methods. Such instances include but are certainly sents multiple distinct neighboring binding pockets and there
not limited to: are fragments which occupancy a single or a number of these
sites taken from one or more crystal structures [211]. The
• studying the electrostatic and dispersion profile of an optimal fragment linking process will result in a compound
X-ray crystal fragment hit, which has potency greater than the sum total of the individ-
• during SBDD and FBDD when there is a requirement ual fragments. This phenomenon is referred to as positive
to rationalize a SAR based on complex interactions, cooperativity [212]. A recent review identified several ex-
amples of fragment linking in the literature [213]. This has
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 11

been demonstrated in the fragment linking of two Hsp90 allows these terms to be extracted from the FMO calculation,
fragments, 1 (IC50 = 1500 M) and 2 (IC50 = 1000 M), to as shown in Fig. (3). Fragment 1 has a larger electrostatic
yield the more potent inhibitor 3 (IC50 = 1.5 M) [214], see contribution to the sum of the PIE than fragment 2, which
Fig. (2). instead has a more dominant dispersion energy contribution.
Hydrogen bonds are highly directional thus when they are
formed between a fragment and a receptor they act as an
anchor to hold the fragment in place. This is in contrast with
fragments which bind through dispersion forces where mul-
tiple binding poses are often possible, as was observed for
fragment 2 where two crystal structures gave two different
binding poses [214]. As well as being very important infor-
mation in fragment linking, this is particularly useful in the
selection of fragments hits on which fragment expan-
sion/evolution will be proposed.
For protein-ligand interactions, particularly in the context
of drug discovery and guiding medicinal chemistry, the most
important pairwise fragment interactions are those between
the ligand and the protein fragments. Subtle changes in
ligand substituents can be studied using FMO in a con-
generic series of compounds which have well resolved crys-
Fig. (2). The complex of fragments 1 (the left part of the ligand in tal structures. This has been demonstrated through the analy-
the center, shown in green) and 2 (the right part of the ligand in the sis of 14 X-ray structures taken from the literature of CDK2
center, shown in cyan) bound together in Hsp90 (PDB ID: 3HZ1) inhibitors [209]. Here, the sum of the PIE for the 14 inhibi-
aligned to the crystal structure (PDB ID: 3HZ5) of the Hsp90 in- tors correlated well (r2 = 0.68) with the free energy of bind-
hibitor 3 (yellow). (The color version of the figure is available in ing as calculated from the measured potencies. This “SAR-
the electronic copy of the article). by-FMO” approach assumes that protein induced conforma-
tional changes upon the ligand modification are kept to a
These results were supported by FMO analysis using minimum and that the only requirement is for proton re-
MP2 and the 6-31G* basis set in combination with PCM Fig. optimization. To speed up the calculations, solvation effects
(3). As calculated by the FMO method, linking fragments 1 are generally ignored. Although solvation and entropic terms
(-8.87 kcal/mol) and 2 (-9.03 kcal/mol) via a propyl linker to can be added in later, it is clear that FMO calculations can be
yield compound 3 (-41.3 kcal/mol) results in an increase in used in this simple manner, where optimizing primarily for
binding energy [214]. In successful fragment linking there is enthalpy is desired in pockets which are not too solvent ex-
a requirement to maintain the thermodynamic profile of the posed. Sum of the PIE derived from FMO calculations can
linking partners. This is analogous to optimizing the thermo- then be used to rank order compounds and to help prioritize
dynamic signature in lead discovery [215]. Fragments which compounds for synthesis.
are stabilized through H-bonds to the receptor (enthalpic In situations where speed is important, the additional
binders) and in addition can accommodate a degree of sub- terms to treat solvation and entropy can be sought from other
stitution are an attractive starting points for fragment expan- means and combined with the enthalpic contribution from
sion [216]. Fragments which are bound through more hydro- FMO by performing partial least squares (PLS) regression on
phobic interactions and tend to have a more entropic ther- the combined terms. This is particularly useful when the in-
modynamic signature are ideal fragments to which an en- dividual terms come from disparate methods, as the terms
thalpic fragment should be linked. This entropic signature is can then be scaled appropriately. QSAR models built using
associated with the hydrophobic effect of ligand binding. PLS can be used as scoring functions to predict ligand po-
Contribution of hydration to the thermodynamics of ligand tency. As an example, the calculation of ligand-binding free
binding has been a topic of considerable interest for recent energies was modeled using the FMO method to describe
years [217, 218]. It is generally understood that the major enthalpic binding term, a polar solvation term described us-
energetic contribution to the hydrophobic effect during the ing Poisson-Boltzmann surface area (PBSA) techniques, a
association of a ligand and its receptor comes from the re- nonpolar solvation term estimated from the solvent-
moval of high energy (unstable) water molecules from the accessible surface area of the ligand and a ligand-based ro-
binding site and essentially entropic in nature. We believe tatable bond count as a measure of conformational ligand
that the ligand/receptor interface with high dispersion inter- entropy [209]. A QSAR model was built using a training set
action term identified by FMO PIEDA is very likely to be derived from the 14 CDK2 inhibitors which were structurally
associated with the presence of these high energy waters and solved. The resulting model was very predictive, with an r2
research to prove this hypothesis is currently ongoing. Hy- of 0.939 and a q2 of 0.896. This model was then used to pre-
drogen bonds have a large electrostatic energy term whereas dict a data set of a further 14 CDK2 inhibitors from the lit-
the dispersion energy term is the dominant attractive compo- erature [219, 220] giving an r2 of the test set of 0.68 and a
nent in hydrophobic interactions. The ratio of the electro- root mean squared error of prediction of 1.00.
static and dispersion energy terms to the sum of the pair in-
teraction energy results is a useful indication of the suitabil- In an alternative procedure, a virtual library of com-
ity of a fragment for hit expansion see Fig. (3). PIEDA [90] pounds can be built around a single crystal structure, and
allow the use of FMO in a predictive fashion, without the
12 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

Fig. (3). Left: PIEDA for fragments 1 and 2. Showing the electrostatic term (Ees= ), the exchange repulsion term (Eex= ), the CT
and mixing terms (Ect+mix= ) and the dispersion term (Edisp= ), energies are in kcal/mol. Right: Table showing the activity
in (μM); dG, the free energy of binding at 298K; FMO+PCM, the binding energy calculated using FMO and the PCM water model;
ES/(D+CT), ratio of the electrostatic term over the sum of the dispersion and CT terms extracted from the PIEDA. (The color version of the
figure is available in the electronic copy of the article).

need for a QSAR model. In this procedure, the size of the In order to demonstrate the usefulness of “SAR-by-
system to be calculated is kept minimum, i.e. typically the FMO” approach we have compiled a small virtual (except
ligand and only the surrounding amino acid residues, in or- compound 13 which was actually synthesized in the work)
der to ensure reasonable throughput. If it is appropriate to do library of 1WCC ligand analogues (Table 1) and their FMO
so, only a part of a ligand, the R-group moiety which is to be interaction energies calculated Fig. (5) so that the effect of
changed during a hit/lead optimization, can be included. This substituent changes on binding free energy can be predicted.
technique is often used for initial fragment hit expansion or Replacing the chloro (entry 1 Table 1) for a fluoro (entry 2),
for optimization of a congeneric series of compounds and proton (entry 3) or methyl group (entry 12) all result in a loss
particularly effective for solvent inaccessible rigid biding in binding energy, similarly, entries 4, 5 and 7. This suggests
pockets as most of the energetic terms not accounted for by that the chloro is necessary for activity. Further changes to
FMO method tend to cancel out. We term this procedure the ring system, pyrazine to pyridine (entry 11, Table 1)
“SAR-by-FMO” and an example of this is presented below. could be demonstrated by FMO to be less favorable changes.
This method can also be useful for scaffold hopping and in However, substituent changes to the amino group could offer
fragment replacement. further gains in potency. Indeed, the authors went on to pre-
pare compound 13 (Table 1) which had reasonable activity
As an example, Wyatt et al. described the fragment-
based drug design against CDK2, starting from a number of (IC50 = 7 μM). Using the “SAR-by-FMO” approach con-
firms this approach with a favorable change in binding en-
fragment hits, which were then crystallized [220]. One of the
ergy. This analysis reveals that there are other scaffolds that
resolved structures was the fragment hit (PDB ID: 1WCC),
could have also been interesting fragment hits for further
entry 1 in (Table 1). This fragment has an activity of 64%
investigation, including entry 6, 9 and 10 (Table 1), all of
inhibition at 1mM. It was of initial interest as there are three
which have increased binding interactions to the receptor
interactions formed between the fragment and the Hinge
region backbone of the CDK2 receptor along residues compared to the 1WCC fragment hit.
Glu81, Phe82 and Leu83. These are depicted in Fig. (4), and FMO can have particular utility in understanding protein-
following the FMO convention of fragmentation are labeled protein interaction (PPI) targets. FMO is extremely powerful
Phe-82, Leu-83 and His-84, respectively. An interaction that in quickly identifying important structural features around a
was not thoroughly examined in the literature is that formed moiety and for quickly establishing virtual SAR to prioritize
between the chloro substituent and the gate-keeper Phe80 compounds prior to synthesis. This is a suitable method for
residue, which contributes to about 5 kcal/mol of energy. At FBDD and for small sub-structural changes which are not in
the MP2/6-31G* level of theory the interaction can be ob- conjugation to a larger parent ligand. As well as fragmenting
served as being a combination of dispersion and CT energy proteins into amino acid monomers, large ligands can also be
terms. This is likely as the chloro group has a positively fragmented across single bonds. This is particularly useful in
charged  hole which is likely to interact with the electron the modeling of peptide mimetics to target protein-protein
rich -system of Phe-80 [221], an interaction which is poorly interactions or peptide substrate receptors, for example, in
represented in MM force fields and often overlooked. Al- the structure-based design of peptide-like, type-1, renin in-
though it is also reasonable to assume that the chlorine atom hibitors to treat primary hypertension [222]. Here, potent in
may be replacing a high energy solvating water molecule vitro inhibitors, like CGP 38’560, mimic the binding mode
which might exist in this part of the biding pocket, the con- of the endogenous renin substrate angiotensinogen, Fig. (6).
tribution from this type of halogen- interaction cannot be These large compounds suffer from a high molecular weight
ignored. The authors attempted to change the chloro sub- and high lipophilicity leading to poor oral absorption and
stituent for alternate functional groups without success and rapid biliary uptake. Rahuel et al. reported the optimization
consequently the series was dropped. of renin inhibitors by truncating the large peptidomimetic to
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 13

Table 1. SAR-by-FMO. Fragment Analysis Using FMO to Calculate the Total Sum of the Pair Interaction Energies (PIEs)

Entry PIE (kcal/mol)


R1 N R2

A R1 R2

1 -21.44 N Cl NH2

2 -20.13 N F NH2

3 -20.23 N H NH2

4 -20.00 C H NH2

5 -21.07 C CH3 NH2

6 -22.53 N Cl H

7 -21.32 N H H

8 -21.13 N Cl OCH3

9 -24.20 N Cl NHCOCH3

10 -24.06
Cl N

11 -21.10 C Cl NH2

12 -21.06 N CH3 NH2

13 -23.03 N Cl NHPh

Fig. (4). FMO analysis of 1WCC. A) PIEDA for Entry 1 Table 1 (PDB ID: 1WCC) showing the electrostatic term (Ees), the exchange repul-
sion term (Eex), the CT and mixing terms (Ect+mix) and the dispersion term (Edisp), energies are in kcal/mol. B) pair interaction energies
(PIEs) for each of the amino acids neighboring the fragment, showing attractive and repulsive energy contributions in kcal/mol. C) Depiction
of the fragment within the CDK2 binding pocket, taken from the PDB ID: 1WCC [10]. (The color version of the figure is available in the
electronic copy of the article).
14 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

Fig. (5). Plot of the difference in pair interaction energies (PIE) from 1WCC. Graph showing the difference in the total sum of the PIEs be-
tween fragment 1 ( Entry 1, Table 1) and 12 other fragments from Table 1. Fragments which have enhanced binding interaction energy as
compared to fragment 1 are shown as blue bars pointing downwards (with negative energy differences). (The color version of the figure is
available in the electronic copy of the article).

Fig. (6). X-ray structures of peptidomimetics and fragmentation scheme used in the FMO analysis. Left: Top: CGP 38’560 (PDB ID: 1RNE);
middle: compound 7 (PDB ID: 2V13); bottom: Aliskiren (PDB ID: 2V0Z). Right: The corresponding fragmentation scheme used during the
FMO calculations. (The color version of the figure is available in the electronic copy of the article).
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 15

improve ligand efficiency and physiochemical properties These few case studies demonstrate how FMO can pro-
resulting in the discovery of Aliskiren [222]. With the avail- vide important information on the nature of protein ligand
ability of the X-ray structure of CGP 38’560 [223] and its interactions to support medicinal chemistry projects. In par-
analogues [222], FMO can be used to examine these struc- ticular the importance of the total sum of the PIE was men-
tures in greater detail. Using FMO, CGP 38’560 can be tioned as a means to establish “SAR-by-FMO”. The analysis
fragmented into 4 components and the contribution of the of fragment contributions via the fragment efficiency, FE,
fragments to the total sum of the PIE can be determined, Fig. metric provides alternative means to prioritize compounds
(6) and (Table 2). An examination of the fragment PIEs re- for further investigation. In addition, the results from FMO
veal that fragment B contributes nearly half of the total PIE. can be presented in a visually meaningful fashion. The appli-
A measure of the fragment efficiency, FE, can be calculated cations demonstrated here were run using FMO as a single
by dividing the fragment PIE by the fragment heavy atom point calculation. Work is ongoing to establish whether
count, FMO/FDD can be developed into a routine method for
SBDD to assess ligand protein interactions to calculate the
Fragment Efficiency = Pair Interaction Energy / Heavy Atom
Count (7) free energy of binding at a greater level of accuracy.

Fragments A, A’ and C have a reduced percentage con- 7. CONCLUSION AND OUTLOOK


tribution to the total PIE, and fragments A and B have low
FEs, indicating inefficient receptor binding for these frag- The FMO and QM/MM methods are reliable computa-
ments. In the course of the development compound 7 was tional tools, and the advances in both hardware and software
identified as having improved oral absorption in vivo [222]. development have made these calculations practically avail-
This compound has an overall larger FE compared to CGP able to end users. It seems that in part the reason for a rela-
38’560 despite the removal of the S2 pocket moiety. This tively minor overall use of QM in drug design is the inertia
compound was fragmented similarly to CGP 38’560, Fig. on the part of computational scientists, and it is our hope to
(6). The marked size reduction of fragment B (FE = 5.71), raise the interest to QM methods in drug discovery commu-
CGP 38’560 (FE = 4.78), and the greatly improved binding nity with this review.
of fragment A to the receptor via the S3sp sub-pocket (FE = Clearly, QM methods offer major advantages over other
4.10), CGP 38’560 (FE = 1.71), contributed to the improved approaches, and the difficulty is in selecting the right method
binding efficiency of this compound Fig. (6). The reduction for a given problem. In many cases, especially, when the
in potency is calculated by FMO to be due to the reduction in dynamic (entropic) factor is of major importance, it is con-
attractive binding for fragment B. Performing the same ceivable that traditional MM approaches combined with mo-
fragmentation on Aliskiren and running the FMO method on lecular dynamics will continue to be used for those studies.
this compound showed how potency was regained in this QM methods have a potential to become a primary tool in
clinical candidate. Here, there was an improvement in the many areas of drug design, especially in designing ligands
PIE, (PIE = 83.95 kcal/mol), together the FE of fragment A with high affinity and specificity, because now one can af-
being retained, to give an overall increase in PIE for the ford to use these more expensive but also more accurate ap-
ligand. proaches in actual drug research.

Table 2. Fragment Efficiency (FE). Examination of the Fragment Contributions to the Total Pair Interaction Energies (PIE) for
the Peptidomimetics CGP 38’560, Compound 7 and Aliskiren. Showing the Activity in μM, the Total PIE and Total FE
As well as the Fragment Contributions to PIE and FE and the Percentage (%) of the Total Contribution to the Ligand
PIE. See Figure 5 for the Fragmentation Scheme

Total PIE (kcal/mol)


Compound Activity (μM) Fragment PIE (kcal/mol) % of total PIE FE
[Total FE]

A -27.40 15.5 1.71

A’ -28.43 16.1 2.84


CGP 38’560 0.001 -176.63 [3.46]
B -85.98 48.7 4.78

C -34.84 19.7 4.98

A -65.62 43.1 4.10

7 0.022 -152.28 [4.61] B -57.11 37.5 5.71

C -29.55 19.4 4.22

A -51.47 26.9 3.43

Aliskiren 0.0006 -191.11 [4.90] B -83.95 43.9 6.00

C -55.69 29.2 5.57


16 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

In this review, we have shown how the methods imple- [10] Xie, W.; Gao, J. Design of a next generation force field: the X-POL
mented in GAMESS program suites can be used in realistic potential. J. Chem. Theory Comput. 2007, 3 (6), 1890-1900.
[11] Zhou, T.; Huang, D. Z.; Caflisch, A. Quantum Mechanical methods
studies of biochemical systems, in particular, FMO and for drug design. Curr. Top. Med. Chem. 2010, 10 (1), 33-45.
QM/MM. We have also briefly introduced various user inter- [12] Levine, I. N. Quantum chemistry. Prentice Hall: 1991; Vol. 171.
faces taking care of the technical issues in setting up input [13] Szabo, A.; Ostlund, N. S. Modern quantum chemistry: introduction
files and visualizing the results. We hope that GAMESS will to advanced electronic structure theory. Dover Publications: 1996.
[14] Cramer, C. J. Essentials of computational chemistry: theories and
find its increasing use in drug research. models. John Wiley & Sons Inc: 2004.
[15] Gan, Z.; Alexeev, Y.; Gordon, M. S.; Kendall, R. A. The parallel
CONFLICT OF INTEREST implementation of a full configuration interaction program. J.
Chem. Phys. 2003, 119, 47.
The authors confirm that this article content has no con- [16] Nakatsuji, H.; Nakashima, H.; Kurokawa, Y.; Ishikawa, A. Solving
flicts of interest. the Schrödinger equation of atoms and molecules without
analytical integration based on the free iterative-complement-
interaction wave function. Phys. Rev. Lett. 2007, 99 (24), 240402.
ACKNOWLEDGEMENTS [17] Koch, W.; Holthausen, M. C. A chemist's guide to density
functional theory. Wiley Online Library: 2001; Vol. 2.
YA and DGF thank their former PhD supervisor Prof. [18] Otsuka, T.; Miyazaki, T.; Ohno, T.; Bowler, D.; Gillan, M.
Mark S. Gordon and a long time mentor Dr. Michael W. Accuracy of order-N density-functional theory calculations on
Schmidt, for their lifetime commitment to GAMESS-US DNA systems using CONQUEST. J. Phys.: Condens. Matter 2008,
development and countless fruitful discussions. YA work is 20, 294201.
[19] Goedecker, S. Linear scaling electronic structure methods. Rev.
supported by the Office of Science of the U.S. Department of Mod. Phys. 1999, 71 (4), 1085.
Energy under contract DE-AC02-06CH11357. DGF thanks [20] Scuseria, G. E. Linear scaling density functional calculations with
Prof. Kazuo Kitaura for many insightful discussions and Gaussian orbitals. J. Phys. Chem. A 1999, 103 (25), 4782-4790.
acknowledges partial financial support from the Next Gen- [21] Gordon, M. S.; Mullin, J. M.; Pruitt, S. R.; Roskop, L. B.;
eration Super Computing Project, Nanoscience Program Slipchenko, L. V.; Boatz, J. A. Accurate methods for large
molecular systems. J. Phys. Chem. B 2009, 113 (29), 9646-9663.
(MEXT, Japan) and Strategic Programs for Innovative Re- [22] Gordon, M. S.; Fedorov, D. G.; Pruitt, S. R.; Slipchenko, L. V.
search (SPIRE, Japan). We also thank Dr. Alex A. Gra- Fragmentation methods: a route to accurate calculations on large
novsky for his comments on Firefly. OI and MM thank Dr. systems. Chem. Rev. 2012, 112 (1), 632-672.
Mark Whittaker and Dr. Richard Law for many inspiring [23] Nikitina, E.; Sulimov, V.; Zayets, V.; Zaitseva, N. Semiempirical
calculations of binding enthalpy for protein–ligand complexes. Int.
comments and support. J. Quantum Chem. 2004, 97 (2), 747-763.
[24] Stewart, J. J. P. Application of the PM6 method to modeling
REFERENCES proteins. J. Mol. Model. 2009, 15 (7), 765-805.
[25] Cheeseman, K. H.; Slater, T. F. An introduction to free radical
[1] Merz, K. M.; Ringe, D.; Reynolds, C. H., Eds., Drug design: biochemistry. Br. Med. Bull. 1993, 49 (3), 481-493.
structure-and ligand-based approaches. Cambridge University [26] Kitaura, K.; Ikeo, E.; Asada, T.; Nakano, T.; Uebayasi, M.
Press: Cambridge, 2010. Fragment molecular orbital method: an approximate computational
[2] Vanommeslaeghe, K.; Hatcher, E.; Acharya, C.; Kundu, S.; Zhong, method for large molecules. Chem. Phys. Lett. 1999, 313 (3-4),
S.; Shim, J.; Darian, E.; Guvench, O.; Lopes, P.; Vorobyov, I.; 701-706.
Mackerell, A. D. J. CHARMM general force field: A force field for [27] Fletcher, G. D.; Fedorov, D. G.; Pruitt, S. R.; Windus, T. L.;
drug-like molecules compatible with the CHARMM all-atom Gordon, M. S. Large-scale MP2 calculations on the Blue Gene
additive biological force fields. J. Comput. Chem. 2010, 31 (4), architecture using the Fragment Molecular Orbital method. J.
671-690. Chem. Theory Comput. 2012, 8, 75-79.
[3] Pearlman, D. A.; Case, D. A.; Caldwell, J. W.; Ross, W. S.; [28] Alexeev, Y.; Mahajan, A.; Leyffer, S.; Fletcher, G. D.; Fedorov, D.
Cheatham, T. E.; DeBolt, S.; Ferguson, D.; Seibel, G.; Kollman, P. G. In Heuristic static load-balancing algorithm applied to the
AMBER, a package of computer programs for applying molecular Fragment Molecular Orbital method, Proceedings of the
mechanics, normal mode analysis, molecular dynamics and free ACM/IEEE Supercomputing 2012 Conference, Salt Lake City,
energy calculations to simulate the structural and energetic IEEE: Salt Lake City, 2012.
properties of molecules. Comput. Phys. Commun. 1995, 91 (1), 1- [29] Andricioaei, I.; Karplus, M. On the calculation of entropy from
41. covariance matrices of the atomic fluctuations. J. Chem. Phys.
[4] Ponder, J. W.; Case, D. A. Force fields for protein simulations. 2001, 115, 6289.
Adv. Protein Chem. 2003, 66, 27-85. [30] Schmidt, M. W.; Baldridge, K. K.; Boatz, J. A.; Elbert, S. T.;
[5] Dupuis, M.; Aida, M.; Kawashima, Y.; Hirao, K. A polarizable Gordon, M. S.; Jensen, J. H.; Koseki, S.; Matsunaga, N.; Nguyen,
mixed Hamiltonian model of electronic structure for micro- K. A.; S., S.; Windus, T. L.; Dupuis, M.; Montgomery, J. A. J.
solvated excited states. I. Energy and gradients formulation and General atomic and molecular electronic structure system. J.
application to formaldehyde 1A2. J. Chem. Phys. 2002, 117, 1242. Comput. Chem. 1993, 14 (11), 1347-1363.
[6] Ponder, J. W.; Wu, C.; Ren, P.; Pande, V. S.; Chodera, J. D.; [31] Guest, M. F.; Bush, I. J.; Van Dam, H. J. J.; Sherwood, P.; Thomas,
Schnieders, M. J.; Haque, I.; Mobley, D. L.; Lambrecht, D. S.; J. M. H.; Van Lenthe, J. H.; Havenith, R. W. A.; Kendrick, J. The
DiStasio Jr, R. A.; Head-Gordon, M.; Clark, G. N. I.; Johnson, M. GAMESS-UK electronic structure package: algorithms,
E.; Head-Gordon, T. Current status of the AMOEBA polarizable developments and applications. Mol. Phys. 2005, 103 (6-8), 719-
force field. J. Phys. Chem. B 2010, 114 (8), 2549-2564. 747.
[7] Lamoureux, G.; MacKerell Jr, A. D.; Roux, B. A simple [32] Gordon, M.S.; Schmidt, M.W. Recent advances in QM and
polarizable model of water based on classical Drude oscillators. J. QM/MM methods. Sloot, P.; Abramson, D.; Bogdanov, A.;
Chem. Phys. 2003, 119, 5185. Gorbachev, Y.; Dongarra, J.; Zomaya, A., Eds. Springer Berlin,
[8] Day, P. N.; Jensen, J. H.; Gordon, M. S.; Webb, S. P.; Stevens, W. 2003; Vol. 2660, pp 75-83.
J.; Krauss, M.; Garmer, D.; Basch, H.; Cohen, D. An effective [33] Gordon, M. S.; Schmidt, M. W. Advances in electronic structure
fragment method for modeling solvent effects in quantum theory: GAMESS a decade later. In Theory and Applications of
mechanical calculations. J. Chem. Phys. 1996, 105 (5), 1968-1986. Computational Chemistry the first forty years, Dykstra, C.;
[9] Ji, C.; Mei, Y.; Zhang, J. Z. H. Developing polarized protein- Frenking, G.; Kim, K.; Scuseria, G., Eds. Elsevier Science:
specific charges for protein dynamics: MD free energy calculation Amsterdam, 2005; pp 1167-1189.
of pKa shifts for Asp26/Asp20 in thioredoxin. Biophys. J. 2008, 95 [34] Computing for Science Ltd. GAMESS-UK. http://www.cfs.dl.ac.uk
(3), 1080-1088. / (Accessed January 16, 2012).
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 17

[35] Granovsky, A. A. Firefly version 7.1.G. http://classic.chem.msu.su/ [56] Ishimura, K.; Kuramoto, K.; Ikuta, Y.; Hyodo, S. MPI/OpenMP
gran/firefly/index.html (Accessed January 16, 2012). hybrid parallel algorithm for Hartree - Fock calculations. J. Chem.
[36] Mark Gordon's Quantum Theory Group: Download GAMESS. Theory Comput. 2010, 6 (4), 1075-1080.
http://www.msg.ameslab.gov/gamess/download.html (Accessed [57] Weltman, J. K.; Skowron, G.; Loriot, G. B. HIV-1 GP120 V3
January 16, 2012). conformational and informational entropies. J. Mol. Model. 2006,
[37] Higashi, M.; Marenich, A. V.; Olson, R. M.; Chamberlin, A. C.; 12 (3), 362-365.
Pu, J.; Kelly, C. P.; Thompson, J. D.; Xidos, J. D.; Li, J.; Zhu, T.; [58] Imamura, A.; Aoki, Y.; Maekawa, K. A theoretical synthesis of
Hawkins, G. D.; Chuang, Y.; Fast, P. L.; Lynch, B. J.; Liotard, D. polymers by using uniform localization of molecular orbitals:
A.; Rinaldi, D.; Gao, J.; Cramer, C. J.; Truhlar, D. G. University of Proposal of an elongation method. J. Chem. Phys. 1991, 95, 5419-
Minnesota: GAMESSPLUS - addon for GAMESS-US. 5431.
http://comp.chem.umn.edu/gamessplus/ (Accessed January 16, [59] Kobayashi, M.; Akama, T.; Nakai, H. Second-order Møller-Plesset
2012). perturbation energy obtained from divide-and-conquer Hartree-
[38] Google Inc. Google groups: GAMESS. http://groups.google.com/ Fock density matrix. J. Chem. Phys. 2006, 125, 204106.
group/gamess (Accessed January 16, 2012). [60] Orimoto, Y.; Gu, F. L.; Imamura, A.; Aoki, Y. Efficient and
[39] Fletcher, G. D.; Schmidt, M. W.; Bode, B. M.; Gordon, M. S. The accurate calculations on the electronic structure of B-type poly
distributed data interface in GAMESS. Comput. Phys. Commun. (dG)•poly (dC) DNA by elongation method: First step toward the
2000, 128 (1-2), 190-200. understanding of the biological properties of aperiodic DNA. J.
[40] Gropp, W.; Lusk, E.; Skjellum, A. Using MPI: portable parallel Chem. Phys. 2007, 126 (21), 215104.
programming with the message passing interface. MIT Press: [61] Xie, P.; Liu, K.; Gu, F.; Aoki, Y. Counter-ion effects of A- and B-
Cambridge, MA, USA, 1999. type poly (dG)·Poly (dC) and poly (dA)· Poly (dT) DNA by
[41] Fletcher, G.D.; Schmidt, M.W.; Gordon, M.S. Developments in elongation method. Int. J. Quantum Chem. 2012, 112 (1), 230–239.
parallel electronic structure theory. Adv. Chem. Phys. 1999, 267- [62] Aoki, Y.; Gu, F. L. An elongation method for large systems toward
294. bio-systems. Phys. Chem. Chem. Phys. 2012, 14 (21), 7640-7668.
[42] Alexeev, Y.; Kendall, R. A.; Gordon, M. S. The distributed data [63] Avramov, P. V.; Fedorov, D. G.; Sorokin, P.; Sakai, S.; Entani, S.;
SCF. Comput. Phys. Commun. 2002, 143 (1), 69-82. Ohtomo, M.; Matsumoto, Y.; Naramoto, H. Intrinsic edge
[43] Alexeev, Y.; Schmidt, M. W.; Windus, T. L.; Gordon, M. S. A asymmetry in narrow zigzag hexagonal heteroatomic nanoribbons
parallel distributed data CPHF algorithm for analytic Hessians. J. causes their subtle uniform curvature. J. Phys. Chem. Lett. 2012, 3,
Comput. Chem. 2007, 28 (10), 1685-1694. 2003-2008.
[44] Fedorov, D. G.; Olson, R. M.; Kitaura, K.; Gordon, M. S.; Koseki, [64] Fedorov, D. G.; Kitaura, K. Extending the power of quantum
S. A new hierarchical parallelization scheme: Generalized chemistry to large systems with the fragment molecular orbital
distributed data interface (GDDI), and an application to the method. J. Phys. Chem. A 2007, 111 (30), 6904-6914.
fragment molecular orbital method (FMO). J. Comput. Chem. [65] Fedorov, D. G.; Nagata, T.; Kitaura, K. Exploring chemistry with
2004, 25 (6), 872-880. the fragment molecular orbital method. Phys. Chem. Chem. Phys.
[45] Stephens, P.; Devlin, F.; Chabalowski, C.; Frisch, M. J. Ab initio 2012, 14, 7562-7577.
calculation of vibrational absorption and circular dichroism spectra [66] Fedorov, D. G.; Kitaura, K. Theoretical development of the
using density functional force fields. J. Phys. Chem. 1994, 98 (45), fragment molecular orbital (FMO) method. In Modern methods for
11623-11627. theoretical physical chemistry of biopolymers, Starikov, E. B.;
[46] Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized gradient Lewis, J. P.; Tanaka, S., Eds. Elsevier: Amsterdam, The
approximation made simple. Phys. Rev. Lett. 1996, 77 (18), 3865- Netherlands, 2006; pp 3-38.
3868. [67] Nakano, T.; Mochizuki, Y.; Fukuzawa, K.; Amari, S.; Tanaka, S.
[47] Yanai, T.; Tew, D. P.; Handy, N. C. A new hybrid exchange- Developments and applications of ABINIT-MP software based on
correlation functional using the Coulomb-attenuating method the fragment molecular orbital method. In Modern methods for
(CAM-B3LYP). Chem. Phys. Lett. 2004, 393 (1-3), 51-57. theoretical physical chemistry of biopolymers, Starikov, E. B.;
[48] Zhao, Y.; Truhlar, D. G. The M06 suite of density functionals for Lewis, J. P.; Tanaka, S., Eds. Elsevier: Amsterdam, The
main group thermochemistry, thermochemical kinetics, Netherlands, 2006; pp 39-52.
noncovalent interactions, excited states, and transition elements: [68] Nagata, T.; Fedorov, D. G.; Kitaura, K. Mathematical formulation
two new functionals and systematic testing of four M06-class of the fragment molecular orbital method. In Linear-Scaling
functionals and 12 other functionals. Theor. Chem. Acc. 2008, 120 Techniques in Computational Chemistry and Physics, Zalesny, R.;
(1), 215-241. Papadopoulos, M. G.; Mezey, P. G.; Leszczynski, J., Eds. 2011; pp
[49] Grimme, S. Accurate description of van der Waals complexes by 17-64.
density functional theory including empirical corrections. J. [69] Fedorov, D.; Kitaura, K., Eds.,The fragment molecular orbital
Comput. Chem. 2004, 25 (12), 1463-1473. method: practical applications to large molecular systems. CRC
[50] Grimme, S. Improved second-order Møller–Plesset perturbation Press: Boca Raton, FL, 2009.
theory by separate scaling of parallel-and antiparallel-spin pair [70] Steinmann, C.; Fedorov, D. G.; Jensen, J. H. Effective Fragment
correlation energies. J. Chem. Phys. 2003, 118, 9095. Molecular Orbital method: a merger of the Effective Fragment
[51] Feyereisen, M.; Fitzgerald, G.; Komornicki, A. Use of approximate Potential and Fragment Molecular Orbital methods. J. Phys. Chem.
integrals in ab initio theory. An application in MP2 energy A 2010, 114, 8705–8712.
calculations. Chem. Phys. Lett. 1993, 208 (5-6), 359-363. [71] Steinmann, C.; Fedorov, D. G.; Jensen, J. H. The Effective
[52] Brooks, B. R.; Brooks III, C.L.; Mackerell Jr, A.; Nilsson, L.; Fragment Molecular Orbital method for fragments connected by
Petrella, R.; Roux, B.; Won, Y.; Archontis, G.; Bartels, C.; covalent bonds. PLoS One 2012, 7 (7), e41117.
Boresch, S.; Caflisch, A.; Caves, L.; Cui, Q.; Dinner, A. R.; Feig, [72] Suenaga, M. Facio: new computational chemistry environment for
M.; Fischer, S.; Gao, J.; Hodoscek, M.; Im, W.; Kuczera, K.; PC GAMESS. J. Comput. Chem. Jpn 2005, 4 (1), 25-32.
Lazaridis, T.; Ma, J.; Ovchinnikov, V.; Paci, E.; Pastor, R. W.; [73] Kitaura, K.; Sugiki, S. I.; Nakano, T.; Komeiji, Y.; Uebayasi, M.
Post, C. B.; Pu, J. Z.; Schaefer, M.; Tidor, B.; Venable, R. M.; Fragment molecular orbital method: analytical energy gradients.
Woodcock, H. L.; Wu, X.; Yang, W.; York, D. M.; Karplus, M. Chem. Phys. Lett. 2001, 336 (1-2), 163-170.
CHARMM: the biomolecular simulation program. J. Comput. [74] Nagata, T.; Fedorov, D. G.; Ishimura, K.; Kitaura, K. Analytic
Chem. 2009, 30 (10), 1545-1614. energy gradient for second-order Møller-Plesset perturbation theory
[53] Ponder, J. W. TINKER: Software tools for molecular design, based on the fragment molecular orbital method. J. Chem. Phys.
Washington University School of Medicine: Saint Louis, MO, 2011, 135, 044110.
2004. [75] Fedorov, D. G.; Ishida, T.; Uebayasi, M.; Kitaura, K. The fragment
[54] Greengard, L.; Rokhlin, V. A fast algorithm for particle molecular orbital method for geometry optimizations of
simulations. J. Comput. Phys. 1987, 73 (2), 325-348. polypeptides and proteins. J. Phys. Chem. A 2007, 111 (14), 2722-
[55] Choi, C. H.; Ivanic, J.; Gordon, M. S.; Ruedenberg, K. Rapid and 2732.
stable determination of rotation matrices between spherical [76] Ishikawa, T.; Yamamoto, N.; Kuwata, K. Partial energy gradient
harmonics by direct recursion. J. Chem. Phys. 1999, 111, 8825. based on the fragment molecular orbital method: Application to
geometry optimization. Chem. Phys. Lett. 2010, 500 (1), 149-154.
18 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

[77] Fedorov, D. G.; Alexeev, Y.; Kitaura, K. Geometry optimization of perturbation theory for the S22 test set. J. Chem. Theory Comput.
the active site of a large system with the Fragment Molecular 2012, 8(8), 2835-2843.
Orbital method. J. Phys. Chem. Lett. 2011, 2, 282-288. [100] Kairys, V.; Jensen, J. H. QM/MM boundaries across covalent
[78] Nakano, T.; Kaminuma, T.; Sato, T.; Fukuzawa, K.; Akiyama, Y.; bonds: a frozen localized molecular orbital-based approach for the
Uebayasi, M.; Kitaura, K. Fragment molecular orbital method: use effective fragment potential method. J. Phys. Chem. A 2000, 104
of approximate electrostatic potential. Chem. Phys. Lett. 2002, 351 (28), 6656-6665.
(5-6), 475-480. [101] Nagata, T.; Fedorov, D. G.; Sawada, T.; Kitaura, K.; Gordon, M. S.
[79] Choi, C. H.; Fedorov, D. G. Reducing scaling of the fragment A combined effective fragment potential–fragment molecular
molecular orbital method using the multipole method. Chem. Phys. orbital method. II. Analytic gradient and application to the
Lett. 2012, 543, 159-165. geometry optimization of solvated tetraglycine and chignolin. J.
[80] Nagata, T.; Fedorov, D. G.; Kitaura, K. Analytic gradient for the Chem. Phys. 2011, 134, 034110.
embedding potential with approximations in the fragment [102] Hu, L. H.; Eliasson, J.; Heimdal, J.; Ryde, U. Do quantum
molecular orbital method. Chem. Phys. Lett. 2012, 544, 87-93. mechanical energies calculated for small models of protein-active
[81] Fedorov, D. G.; Kitaura, K. Second order Møller-Plesset sites converge? J. Phys. Chem. A 2009, 113 (43), 11793-11800.
perturbation theory based upon the fragment molecular orbital [103] Friesner, R. A.; Guallar, V. Ab initio quantum chemical and mixed
method. J. Chem. Phys. 2004, 121, 2483-2490. quantum mechanics/molecular mechanics (QM/MM) methods for
[82] Fedorov, D. G.; Kitaura, K. On the accuracy of the 3-body studying enzymatic catalysis. Annu. Rev. Phys. Chem. 2005, 56,
fragment molecular orbital method (FMO) applied to density 389-427.
functional theory. Chem. Phys. Lett. 2004, 389 (1-3), 129-134. [104] Gao, J.; Thompson, M. A., Eds.,Combined quantum mechanical
[83] Pruitt, S. R.; Fedorov, D. G.; Kitaura, K.; Gordon, M. S. Open-shell and molecular mechanical methods. American Chemical Society:
formulation of the Fragment Molecular Orbital method. J. Chem. 1998; Vol. 712.
Theory Comput. 2010, 6 (1), 1-5. [105] Lin, H.; Truhlar, D. G. QM/MM: what have we learned, where are
[84] Pruitt, S. R.; Fedorov, D. G.; Gordon, M. S. Geometry we, and where do we go from here? Theor. Chem. Acc. 2007, 117
optimizations of open-shell systems with the Fragment Molecular (2), 185-199.
Orbital method. J. Phys. Chem. A 2012, 116, 4965-4974. [106] Field, M. J.; Bash, P. A.; Karplus, M. A combined quantum
[85] Nakata, H.; Fedorov, D. G.; Nagata, T.; Yokojima, S.; Ogata, K.; mechanical and molecular mechanical potential for molecular
Kitaura, K.; Nakamura, S. Unrestricted Hartree-Fock based on the dynamics simulations. J. Comput. Chem. 1990, 11 (6), 700-733.
Fragment Molecular Orbital method: energy and its analytic [107] Lee, Y. S.; Hodoscek, M.; Brooks, B. R.; Kador, P. F. Catalytic
gradient. J. Chem. Phys. 2012, 137, 044110. mechanism of aldose reductase studied by the combined potentials
[86] Fedorov, D. G.; Kitaura, K. The importance of three-body terms in of quantum mechanics and molecular mechanics. Biophys. Chem.
the fragment molecular orbital method. J. Chem. Phys. 2004, 120, 1998, 70 (3), 203-216.
6832-6840. [108] Lyne, P. D.; Hodoscek, M.; Karplus, M. A hybrid QM-MM
[87] Nakano, T.; Mochizuki, Y.; Yamashita, K.; Watanabe, C.; potential employing Hartree-Fock or density functional methods in
Fukuzawa, K.; Segawa, K.; Okiyama, Y.; Tsukamoto, T.; Tanaka, the quantum region. J. Phys. Chem. A 1999, 103 (18), 3462-3471.
S. Development of the four-body corrected fragment molecular [109] Théry, V.; Rinaldi, D.; Rivail, J. L.; Maigret, B.; Ferenczy, G. G.
orbital (FMO4) method. Chem. Phys. Lett. 2012, 523, 128-133. Quantum mechanical computations on very large molecular
[88] Kitaura, K.; Morokuma, K. A new energy decomposition scheme systems: The local self-consistent field method. J. Comput. Chem.
for molecular interactions within the Hartree-Fock approximation. 1994, 15 (3), 269-282.
Int. J. Quantum Chem. 1976, 10 (2), 325-340. [110] Cui, Q.; Karplus, M. Molecular properties from combined QM/MM
[89] Chen, W.; Gordon, M. S. Energy decomposition analyses for methods. I. Analytical second derivative and vibrational
many-body interaction and applications to water complexes. J. calculations. J. Chem. Phys. 2000, 112, 1133.
Phys. Chem. 1996, 100 (34), 14316-14328. [111] Maseras, F.; Morokuma, K. IMOMM: A new integrated ab initio
[90] Fedorov, D. G.; Kitaura, K. Pair interaction energy decomposition molecular mechanics geometry optimization scheme of equilibrium
analysis. J. Comput. Chem. 2007, 28 (1), 222-237. structures and transition states. J. Comput. Chem. 1995, 16 (9),
[91] Fedorov, D. G.; Kitaura, K. Energy Decomposition Analysis in 1170-1179.
solution based on the Fragment Molecular Orbital Method. J. Phys. [112] Shoemaker, J. R.; Burggraf, L. W.; Gordon, M. S. SIMOMM: An
Chem. A 2012, 116, 704-719. integrated molecular orbital/molecular mechanics optimization
[92] Jorgensen, W. L.; Tirado-Rives, J. The OPLS [optimized potentials scheme for surfaces. J. Phys. Chem. A 1999, 103 (17), 3245-3251.
for liquid simulations] potential functions for proteins, energy [113] Grigorenko, B. L.; Nemukhin, A. V.; Topol, I. A.; Burt, S. K.
minimizations for crystals of cyclic peptides and crambin. J. Am. Modeling of biomolecular systems with the quantum mechanical
Chem. Soc. 1988, 110 (6), 1657-1666. and molecular mechanical method based on the effective fragment
[93] Halgren, T. A. Merck molecular force field. I. Basis, form, scope, potential technique: Proposal of flexible fragments. J. Phys. Chem.
parameterization, and performance of MMFF94. J. Comput. Chem. A 2002, 106 (44), 10663-10672.
1996, 17 (5-6), 490-519. [114] Nemukhin, A. V.; Grigorenko, B. L.; Topol, I. A.; Burt, S. K.
[94] Gordon, M. S.; Slipchenko, L.; Li, H.; Jensen, J. H. In The effective Flexible effective fragment QM/MM method: Validation through
fragment potential: A general method for predicting intermolecular the challenging tests. J. Comput. Chem. 2003, 24 (12), 1410-1420.
interactions, Annual Reports in Computational Chemistry, [115] Li, H.; Thellamurege, N. M.; Si, D. QuanPol: A QM/MM program
Wheeler, R. A., Ed. American Chemical Society: 2007; pp 177- for high-level electronic structure methods and polarizable force
193. field, to be submitted.
[95] Webb, S. P.; Gordon, M. S. Solvation of the Menshutkin reaction: [116] ChemShell, a Computational Chemistry Shell. http://www.chem
A rigorous test of the effective fragment method. J. Phys. Chem. A shell.org (Accessed January 16, 2012).
1999, 103 (9), 1265-1273. [117] Lin, H.; Zhang, Y.; Truhlar, D. G. QMMM; Version 1.3, University
[96] Adamovic, I.; Gordon, M. S. Solvent effects on the SN2 reaction: of Minnesota: Minneapolis, 2006.
Application of the density functional theory-based effective [118] Orozco, M.; Alhambra, C.; Barril, X.; López, J. M.; Busquets, M.
fragment potential method. J. Phys. Chem. A 2005, 109 (8), 1629- A.; Luque, F. J. Theoretical methods for the representation of
1636. solvent. J. Mol. Model. 1996, 2 (1), 1-15.
[97] Adamovic, I.; Gordon, M. S. Dynamic polarizability, dispersion [119] Li, H.; Gordon, M. S. Polarization energy gradients in combined
coefficient C6 and dispersion energy in the effective fragment quantum mechanics, effective fragment potential, and polarizable
potential method. Mol. Phys. 2005, 103 (2-3), 379-387. continuum model calculations. J. Chem. Phys. 2007, 126, 124112.
[98] Smith, Q. A.; Gordon, M. S.; Slipchenko, L. V. Effective Fragment [120] Miertus, S.; Scrocco, E.; Tomasi, J. Electrostatic interaction of a
Potential study of the interaction of DNA bases. J. Phys. Chem. A solute with a continuum. A direct utilizaion of AB initio molecular
2011, 115 (41), 11269-11276. potentials for the prevision of solvent effects. Chem. Phys. 1981, 55
[99] Flick, J. C.; Kosenkov, D.; Hohenstein, E. G.; Sherrill, C. D.; (1), 117-129.
Slipchenko, L. V. Accurate prediction of non-covalent interaction [121] Onsager, L. Electric moments of molecules in liquids. J. Am.
energies with the Effective Fragment Potential method: Chem. Soc. 1936, 58 (8), 1486-1493.
Comparison of energy components to symmetry-adapted
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 19

[122] Klamt, A.; Schüürmann, G. COSMO: a new approach to dielectric [147] Podolyan, Y.; Leszczynski, J. MaSK: A visualization tool for
screening in solvents with explicit expressions for the screening teaching and research in computational chemistry. Int. J. Quantum
energy and its gradient. J. Chem. Soc., Perkin Trans. 2 1993, (5), Chem. 2009, 109 (1), 8-16.
799-805. [148] Allouche, A. R. Gabedit - A graphical user interface for
[123] Chipman, D. M. Charge penetration in dielectric models of computational chemistry softwares. J. Comput. Chem. 2011, 32 (1),
solvation. J. Chem. Phys. 1997, 106 (24), 10194-10206. 174-182.
[124] Chamberlin, A. C.; Cramer, C. J.; Truhlar, D. G. Predicting [149] Sherwood, P.; van Dam, H. J. J.; Thomas, J. The CCP1 GUI
aqueous free energies of solvation as functions of temperature. J. Project. http://www.cse.scitech.ac.uk/ccg/software/ccp1gui/ (Acce
Phys. Chem. B 2006, 110 (11), 5665-5675. ssed January 16, 2012).
[125] Van Duijnen, P. T.; de Vries, A. H. Direct reaction field force field: [150] Anderson, W. P. A Graphical User Interface for PC GAMESS. J.
A consistent way to connect and combine quantum-chemical and Chem. Educ. 2003, 80 (8), 968.
classical descriptions of molecules. Int. J. Quantum Chem. 1996, [151] Agile Molecule: Ascalaph Quantum. http://www.biomolecular-
60 (6), 1111-1132. modeling.com/Ascalaph/Ascalaph_Quantum.html (Accessed Jan
[126] Barone, V.; Cossi, M.; Tomasi, J. A new definition of cavities for uary 16, 2012).
the computation of solvation free energies by the polarizable [152] Nakamura, S. FMO script for drug discovery software package
continuum model. J. Chem. Phys. 1997, 107, 3210. MOE. http://svl.chemcomp.com/ (Accessed January 16, 2012).
[127] Cui, Q. Combining implicit solvation models with hybrid quantum [153] Jmol: an open-source Java viewer for chemical structures in 3D.
mechanical/molecular mechanical methods: A critical test with http://www.jmol.org/ (Accessed January 16, 2012).
glycine. J. Chem. Phys. 2002, 117, 4720. [154] Schmidt, J.; Polik, W. WebMO, LLC: WebMO Pro. http://www.
[128] Vreven, T.; Mennucci, B.; da Silva, C. O.; Morokuma, K.; Tomasi, webmo.net/ (Accessed January 16, 2012).
J. The ONIOM-PCM method: Combining the hybrid molecular [155] Tanford, C.; Kirkwood, J. G. Theory of protein titration curves. I.
orbital method and the polarizable continuum model for solvation. General equations for impenetrable spheres. J. Am. Chem. Soc.
Application to the geometry and properties of a merocyanine in 1957, 79 (20), 5333-5339.
solution. J. Chem. Phys. 2001, 115, 62. [156] Hoogstraten, C. G.; Choe, S.; Westler, W. M.; Markley, J. L.
[129] Bandyopadhyay, P.; Gordon, M. S.; Mennucci, B.; Tomasi, J. An Comparison of the accuracy of protein solution structures derived
integrated effective fragment - polarizable continuum approach to from conventional and network-edited NOESY data. Protein Sci.
solvation: Theory and application to glycine. J. Chem. Phys. 2002, 1995, 4 (11), 2289-2299.
116, 5023. [157] Ogino, T.; Croll, D. H.; Kato, I.; Markley, J. L. Properties of
[130] Li, H.; Pomelli, C. S.; Jensen, J. H. Continuum solvation of large conserved amino acid residues in tandem homologous protein
molecules described by QM/MM: a semi-iterative implementation domains. Proton NMR studies of the histidines of chicken
of the PCM/EFP interface. Theor. Chem. Acc. 2003, 109 (2), 71-84. ovomucoid. Biochem. 1982, 21 (14), 3452-3460.
[131] Li, H.; Robertson, A. D.; Jensen, J. H. The determinants of [158] Jensen, J. H.; Li, H.; Robertson, A. D.; Molina, P. A. Prediction
carboxyl pKa values in turkey ovomucoid third domain. Proteins: and rationalization of protein pKa values using QM and QM/MM
Struct., Funct., Bioinf. 2004, 55 (3), 689-704. methods. J. Phys. Chem. A 2005, 109 (30), 6634-6643.
[132] Fedorov, D. G.; Kitaura, K.; Li, H.; Jensen, J. H.; Gordon, M. S. [159] Naor, M. M.; Jensen, J. H. Determinants of cysteine pKa values in
The polarizable continuum model (PCM) interfaced with the creatine kinase and 1-antitrypsin. Proteins: Struct., Funct., Bioinf.
fragment molecular orbital method (FMO). J. Comput. Chem. 2004, 57 (4), 799-803.
2006, 27 (8), 976-985. [160] Porter, M. A.; Hall, J. R.; Locke, J. C.; Jensen, J. H.; Molina, P. A.
[133] Li, H.; Fedorov, D. G.; Nagata, T.; Kitaura, K.; Jensen, J. H.; Hydrogen bonding is the prime determinant of carboxyl pKa values
Gordon, M. S. Energy gradients in combined fragment molecular at the N-termini of -helices. Proteins: Struct., Funct., Bioinf.
orbital and polarizable continuum model (FMO/PCM) calculation. 2006, 63 (3), 621-635.
J. Comput. Chem. 2010, 31 (4), 778-790. [161] Wang, P. F.; Flynn, A. J.; Naor, M. M.; Jensen, J. H.; Cui, G.; Merz
[134] Nagata, T.; Fedorov, D.; Li, H.; Kitaura, K. Analytic gradient for Jr, K. M.; Kenyon, G. L.; McLeish, M. J. Exploring the role of the
second order Møller-Plesset perturbation theory with the active site cysteine in human muscle creatine kinase. Biochem.
Polarizable Continuum Model based on the Fragment molecular 2006, 45 (38), 11464-11472.
Orbital method. J. Chem. Phys. 2012, 136, 204112. [162] Powers, N.; Jensen, J. H. Chemically accurate protein structures:
[135] Bode, B. M.; Gordon, M. S. MacMolPlt: a graphical user interface Validation of protein NMR structures by comparison of measured
for GAMESS. J. Mol. Graphics Modell. 1998, 16 (3), 133-138. and predicted pKa values. J. Biomol. NMR 2006, 35 (1), 39-51.
[136] Bode, B. M. MacMolPlt version 7.4.3. http://www.scl.ameslab.gov/ [163] Li, H.; Robertson, A. D.; Jensen, J. H. Very fast empirical
MacMolPlt/#Downloading (Accessed January 16, 2012). prediction and rationalization of protein pKa values. Proteins:
[137] FMOutil 2.1. http://staff.aist.go.jp/d.g.fedorov/fmo/fmoutil.html Struct., Funct., Bioinf. 2005, 61 (4), 704-721.
(Accessed January 16, 2012). [164] Jensen Research Group, Department of Chemistry, University of
[138] Alexeev, Y. FMO portal: Web interface for FMOtools. http:// Copenhagen: PROPKA. http://propka.ki.ku.dk/ (Accessed January
www.fmo-portal.info (Accessed January 16, 2012). 16, 2012).
[139] Steinmann, C.; Ibsen, M. W.; Hansen, A. S.; Jensen, J. H. FragIt: a [165] Bas, D. C.; Rogers, D. M.; Jensen, J. H. Very fast prediction and
tool to prepare input files for fragment based quantum chemical rationalization of pKa values for protein–ligand complexes.
calculations, PLoS ONE 2012, 7(9), e44480. Proteins: Struct., Funct., Bioinf. 2008, 73 (3), 765-783.
[140] Steinmann, C., Department of Chemistry, University of Copenha [166] Li, G.; Cui, Q. pKa Calculations with QM/MM Free Energy
gen: FragIt. http://www.fragit.org (Accessed January 16, 2012). Perturbations. J. Phys. Chem. B 2003, 107 (51), 14521-14528.
[141] Senda N. Tencube Institute Ltd.: Winmostar. http://winmostar.com/ [167] Sham, Y. Y.; Chu, Z. T.; Warshel, A. Consistent calculations of
new/en/ (Accessed January 16, 2012). pKas of ionizable residues in proteins: semi-microscopic and
[142] Senda, N. Development of molecular calculation support system microscopic approaches. J. Phys. Chem. B 1997, 101 (22), 4458-
“Winmostar”. Idemitsu Technical Report 2006, 49 (1), 106-111. 4472.
[143] Avogadro: an open-source molecular builder and visualization tool. [168] Olsson, M. H. M.; Hong, G.; Warshel, A. Frozen density functional
Version 1.03. http://avogadro.openmolecules.net (Accessed Jan free energy simulations of redox proteins: computational studies of
uary 16, 2012). the reduction potential of plastocyanin and rusticyanin. J. Am.
[144] Schaftenaar, G.; Noordik, J. H. Molden: a pre-and post-processing Chem. Soc. 2003, 125 (17), 5025-5039.
program for molecular and electronic structures. J. Comput. Aided [169] Streitwieser, A.; Heathcock, C. H.; Kosower, E. M. Introduction to
Mol. Des. 2000, 14 (2), 123-134. organic chemistry. Macmillan New York: 1992.
[145] Zhurko, G. A.; Zhurko, D. A. ChemCraft: Tool for treatment of [170] Elstner, M.; Porezag, D.; Jungnickel, G.; Elsner, J.; Haugk, M.;
chemical data. http://www.chemcraftprog.com/ (Accessed January Frauenheim, T.; Suhai, S.; Seifert, G. Self-consistent-charge
16, 2012). density-functional tight-binding method for simulations of complex
[146] Fleming, S.; Rohl, A. GDIS: a visualization program for molecular materials properties. Phys. Rev. B 1998, 58 (11), 7260.
and periodic systems. Z. Kristallogr. 2005, 220 (5/6/2005), 580- [171] Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.;
584. Klein, M. L. Comparison of simple potential functions for
simulating liquid water. J. Chem. Phys. 1983, 79, 926.
20 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 Alexeev et al.

[172] Sawada, T.; Fedorov, D. G.; Kitaura, K. Structural and interaction using descriptors derived from molecular dynamics and molecular
analysis of helical heparin oligosaccharides with the fragment orbital calculations. Curr. Comput.-Aided Drug Des. 2009, 5 (1),
molecular orbital method. Int. J. Quantum Chem. 2009, 109 (9), 38-55.
2033-2045. [193] Hitaoka, S.; Matoba, H.; Harada, M.; Yoshida, T.; Tsuji, D.;
[173] Fujimura, K.; Sasabuchi, Y. The role of fluorine atoms in a Hirokawa, T.; Itoh, K.; Chuman, H. Correlation analyses on
fluorinated prostaglandin agonist. ChemMedChem 2010, 5 (8), binding affinity of sialic acid analogues and antiinfluenza drugs
1254-1257. with human neuraminidase using ab Initio MO calculations on their
[174] Goodsell, D. S.; Morris, G. M.; Olson, A. J. Automated docking of complex structures-LERE-QSAR analysis (IV). J. Chem. Inf.
flexible ligands: applications of AutoDock. J. Mol. Recognit. 1996, Model. 2011, 51, 2706-2716.
9 (1), 1-5. [194] Ishida, T.; Fedorov, D. G.; Kitaura, K. All electron quantum
[175] Trott, O.; Olson, A. J. AutoDock Vina: improving the speed and chemical calculation of the entire enzyme system confirms a
accuracy of docking with a new scoring function, efficient collective catalytic device in the chorismate mutase reaction. J.
optimization, and multithreading. J. Comput. Chem. 2010, 31 (2), Phys. Chem. B 2006, 110 (3), 1457-1463.
455-461. [195] Watanabe, T.; Inadomi, Y.; Fukuzawa, K.; Nakano, T.; Tanaka, S.;
[176] Kuntz, I. D.; Blaney, J. M.; Oatley, S. J.; Langridge, R.; Ferrin, T. Nilsson, L.; Nagashima, U. DNA and estrogen receptor interaction
E. A geometric approach to macromolecule-ligand interactions. J. revealed by fragment molecular orbital calculations. J. Phys. Chem.
Mol. Biol. 1982, 161 (2), 269-288. B 2007, 111 (32), 9621-9627.
[177] Jones, G.; Willett, P.; Glen, R. C. Molecular recognition of receptor [196] Taguchi, N.; Mochizuki, Y.; Nakano, T.; Amari, S.; Fukuzawa, K.;
sites using a genetic algorithm with a description of desolvation. J. Ishikawa, T.; Sakurai, M.; Tanaka, S. Fragment molecular orbital
Mol. Biol. 1995, 245 (1), 43-53. calculations on red fluorescent proteins (DsRed and mFruits). J.
[178] Jones, G.; Willett, P.; Glen, R. C.; Leach, A. R.; Taylor, R. Phys. Chem. B 2009, 113 (4), 1153-1161.
Development and validation of a genetic algorithm for flexible [197] James, W. H.; Buchanan, E. G.; Mueller, C. W.; Dean, J. C.;
docking. J. Mol. Biol. 1997, 267, 727–748. Kosenkov, D.; Slipchenko, L. V.; Guo, L.; Reidenbach, A. G.;
[179] Rarey, M.; Kramer, B.; Lengauer, T.; Klebe, G. A fast flexible Gellman, S. H.; Zwier, T. S. Evolution of amide stacking in larger
docking method using an incremental construction algorithm. J. -peptides: Triamide h-bonded cycles. J. Phys. Chem. A 2011, 115,
Mol. Biol. 1996, 261, 470-489. 13783-13798.
[180] McGann, M. FRED pose prediction and virtual screening accuracy. [198] Nemoto, T.; Fedorov, D. G.; Uebayasi, M.; Kanazawa, K.; Kitaura,
J. Chem. Inf. Model. 2011, 51, 578-596. K.; Komeiji, Y. Ab initio fragment molecular orbital (FMO)
[181] Jain, A. N. Surflex: fully automatic flexible molecular docking method applied to analysis of the ligand–protein interaction in a
using a molecular similarity-based search engine. J. Med. Chem. pheromone-binding protein. Comput. Biol. Chem. 2005, 29 (6),
2003, 46, 499-511. 434-439.
[182] Friesner, R. A.; Banks, J. L.; Murphy, R. B.; Halgren, T. A.; Klicic, [199] Nakanishi, I.; Fedorov, D. G.; Kitaura, K. Molecular recognition
J. J.; Daniel, T.; Repasky, M. P.; Knoll, E. H.; Shelley, M.; Perry, J. mechanism of FK506 binding protein: An all-electron fragment
K.; Shaw, D. E.; Francis, P.; Shenkin, P. S. Glide: a new approach molecular orbital study. Proteins: Struct., Funct., Bioinf. 2007, 68
for rapid, accurate docking and scoring. 1. Method and assessment (1), 145-158.
of docking accuracy. J. Med. Chem. 2004, 47 (7), 1739-1749. [200] Sawada, T.; Fedorov, D. G.; Kitaura, K. Role of the key mutation
[183] Friesner, R. A.; Murphy, R. B.; Repasky, M. P.; Frye, L. L.; in the selective binding of avian and human influenza
Greenwood, J. R.; Halgren, T. A.; Sanschagrin, P. C.; Mainz, D. T. hemagglutinin to sialosides revealed by quantum-mechanical
Extra precision glide: docking and scoring incorporating a model of calculations. J. Am. Chem. Soc. 2010, 132, 16862–16872.
hydrophobic enclosure for protein-ligand complexes. J. Med. [201] Sawada, T.; Fedorov, D. G.; Kitaura, K. Binding of influenza A
Chem. 2006, 49 (21), 6177-6196. virus hemagglutinin to the sialoside receptor Is not controlled by
[184] Abagyan, R.; Totrov, M.; Kuznetsov, D. ICM - a new method for the homotropic allosteric effect. J. Phys. Chem. B 2010, 114,
protein modeling and design: applications to docking and structure 15700–15705.
prediction from the distorted native conformation. J. Comput. [202] Murata, K.; Fedorov, D. G.; Nakanishi, I.; Kitaura, K. Cluster
Chem. 1994, 15 (5), 488-506. hydration model for binding energy calculations of proteinligand
[185] Li, Y. L.; Mei, Y.; Zhang, D. W.; Xie, D. Q.; Zhang, J. Z. H. complexes. J. Phys. Chem. B 2008, 113 (3), 809-817.
Structure and dynamics of a dizinc metalloprotein: Effect of charge [203] Ohno, K.; Mori, K.; Orita, M.; Takeuchi, M. Computational
transfer and polarization. J. Phys. Chem. B 2011, 98 (115), 10154– insights into binding of bisphosphates to farnesyl pyrophosphate
10162. synthase. Curr. Med. Chem. 2011, 18 (2), 220-233.
[186] Peters, M. B.; Raha, K.; Merz, K. M. Quantum mechanics in [204] He, X.; Fusti-Molnar, L.; Cui, G.; Merz Jr, K. M. Importance of
structure-based drug design. Current Opinion in Drug Discovery dispersion and electron correlation in ab initio protein folding. J.
and Development 2006, 9 (3), 370-379. Phys. Chem. B 2009, 113 (15), 5290-5300.
[187] Raha, K.; Peters, M. B.; Wang, B.; Yu, N.; Wollacott, A. M.; [205] Faver, J. C.; Zheng, Z.; Merz, K. M. Model for the fast estimation
Westerhoff, L. M.; Merz Jr, K. M. The role of quantum mechanics of basis set superposition error in biomolecular systems. J. Chem.
in structure-based drug design. Drug Discov. Today 2007, 12 (17- Phys. 2011, 135, 144110.
18), 725-731. [206] Prime, M. E.; Andersen, O. A.; Barker, J. J.; Brooks, M. A.; Cheng,
[188] Faver, J. C.; Benson, M. L.; He, X.; Roberts, B. P.; Wang, B.; R. K. Y.; Toogood-Johnson, I.; Courtney, S. M.; Brookfield, F. A.;
Marshall, M. S.; Kennedy, M. R.; Sherrill, C. D.; Merz Jr, K. M. Yarnold, C. J.; Marston, R. W.; Johnson, P. D.; Johnsen, S. F.;
Formal estimation of errors in computed absolute interaction Palfrey, J. J.; Vaidya, D.; Erfan, S.; Ichihara, O.; Felicetti, B.;
energies of protein - ligand complexes. J. Chem. Theory Comput. Palan, S.; Pedret-Dunn, A.; Schaertl, S.; Sternberger, I.; Ebneth, A.;
2011, 7, 790-797. Scheel, A.; Winkler, D.; Toledo-Sherman, L.; Beconi, M.;
[189] Ryde, U.; Greco, C.; De Gioia, L. Quantum refinement of [FeFe] Macdonald, D.; Muñoz-Sanjuan, I.; Dominguez, C.; Wityak, J.
hydrogenase indicates a dithiomethylamine ligand. J. Am. Chem. Discovery and structure–activity relationship of potent and
Soc. 2010, 132 (13), 4512-4513. selective covalent inhibitors of transglutaminase 2 for Huntington’s
[190] Ozawa, T.; Okazaki, K.; Kitaura, K. CH/ hydrogen bonds play a disease. J. Med. Chem. 2012, 55 (3), 1021-1046.
role in ligand recognition and equilibrium between active and [207] Zhang, Q.; Yang, J.; Liang, K.; Feng, L.; Li, S.; Wan, J.; Xu, X.;
inactive states of the 2 adrenergic receptor: an ab initio fragment Yang, G.; Liu, D.; Yang, S. Binding interaction analysis of the
molecular orbital (FMO) study. Biorg. Med. Chem. 2011, 19, active site and its inhibitors for neuraminidase (N1 subtype) of
5231–5237. human influenza virus by the integration of molecular docking,
[191] Amari, S.; Aizawa, M.; Zhang, J.; Fukuzawa, K.; Mochizuki, Y.; FMO calculation and 3D-QSAR CoMFA modeling. J. Chem. Inf.
Iwasawa, Y.; Nakata, K.; Chuman, H.; Nakano, T. VISCANA: Model. 2008, 48 (9), 1802-1812.
visualized cluster analysis of protein-ligand interaction based on [208] Zhang, Q.; Yu, C.; Min, J.; Wang, Y.; He, J.; Yu, Z. Rational
the ab initio fragment molecular orbital method for virtual ligand questing for potential novel inhibitors of FabK from Streptococcus
screening. J. Chem. Inf. Model. 2006, 46 (1), 221-230. pneumoniae by combining FMO calculation, CoMFA 3D-QSAR
[192] Yoshida, T.; Fujita, T.; Chuman, H. Novel quantitative structure- modeling and virtual screening. J. Mol. Model. 2011, 17 (6), 1483-
activity studies of HIV-1 protease inhibitors of the cyclic urea type 1492.
GAMESS for Drug Research Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 18 21

[209] Mazanetz, M. P.; Ichihara, O.; Law, R. J.; Whittaker, M. Prediction [217] Abel, R.; Young, T.; Farid, R.; Berne, B. J.; Friesner, R. A. Role of
of cyclin-dependent kinase 2 inhibitor potency using the fragment the active-site solvent in the thermodynamics of factor Xa ligand
molecular orbital method. J. Cheminform. 2011, 3 (1), 2. binding. J. Am. Chem. Soc. 2008, 130 (9), 2817-2831.
[210] Hitaoka, S.; Harada, M.; Yoshida, T.; Chuman, H. Correlation [218] Huggins, D. J.; Sherman, W.; Tidor, B. Rational approaches to
analyses on binding affinity of sialic acid analogues with influenza improving selectivity in drug design. J. Med. Chem. 2012, 55 (4),
virus Neuraminidase-1 using ab Initio MO calculations on their 1424–1444.
complex structures. J. Chem. Inf. Model. 2010, 50 (10), 1796-1805. [219] Congreve, M.; Chessari, G.; Tisi, D.; Woodhead, A. J. Recent
[211] Neumann, L.; Von König, K.; Ullmann, D. HTS reporter developments in fragment-based drug discovery. J. Med. Chem.
displacement assay for fragment screening and fragment evolution 2008, 51 (13), 3661-3680.
toward leads with optimized binding kinetics, binding selectivity, [220] Wyatt, P. G.; Woodhead, A. J.; Berdini, V.; Boulstridge, J. A.;
and thermodynamic signature. Methods Enzymol. 2011, 493, 299- Carr, M. G.; Cross, D. M.; Davis, D. J.; Devine, L. A.; Early, T. R.;
320. Feltell, R. E. Identification of N-(4-piperidinyl)-4-(2, 6-dichloro
[212] Williams, D. H.; Stephens, E.; O'Brien, D. P.; Zhou, M. benzoylamino)-1 H-pyrazole-3-carboxamide (AT7519), a novel
Understanding noncovalent interactions: Ligand binding energy cyclin dependent kinase inhibitor using fragment-based X-Ray
and catalytic efficiency from ligand-induced reductions in motion crystallography and structure based drug design. J. Med. Chem.
within receptors and enzymes. Angew. Chem. Int. Ed. 2004, 43 2008, 51 (16), 4986-4999.
(48), 6596-6616. [221] Imai, Y. N.; Inoue, Y.; Nakanishi, I.; Kitaura, K. Cl– Interactions
[213] Ichihara, O.; Barker, J.; Law, R. J.; Whittaker, M. Compound in protein–ligand complexes. QSAR Comb. Sci. 2009, 28 (8), 869-
design by fragment-linking. Mol. Inform. 2011, 30 (4), 298-306. 873.
[214] Barker, J. J.; Barker, O.; Courtney, S. M.; Gardiner, M.; [222] Rahuel, J.; Rasetti, V.; Maibaum, J.; Rüeger, H.; Göschke, R.;
Hesterkamp, T.; Ichihara, O.; Mather, O.; Montalbetti, C. A. G. N.; Cohen, N.; Stutz, S.; Cumin, F.; Fuhrer, W.; Wood, J. Structure-
Müller, A.; Varasi, M. Discovery of a novel Hsp90 inhibitor by based drug design: The discovery of novel nonpeptide orally active
fragment linking. ChemMedChem 2010, 5 (10), 1697-1700. inhibitors of human renin. Chem. Biol. 2000, 7 (7), 493-504.
[215] Ferenczy, G. G.; Keseru, G. M. Thermodynamics guided lead [223] Rahuel, J.; Priestle, J. P.; Grütter, M. G. The crystal structures of
discovery and optimization. Drug Discov. Today 2010, 15 (21-22), recombinant glycosylated human renin alone and in complex with a
919-932. transition state analog inhibitor. J. Struct. Biol. 1991, 107 (3), 227-
[216] Ferenczy, G. G.; Keseru, G. M. Thermodynamics of fragment 236.
binding. J. Chem. Inf. Model. 2012, 52 (4), 1039-1045.

Received: May 11, 2012 Revised: July 30, 2012 Accepted: September 06, 2012

View publication stats

You might also like