Consideration of Molecular Weight during Compound Selection in

J. Chem. Inf. Comput. Sci. 2003, 43, 267-272
267
Consideration of Molecular Weight during Compound Selection in Virtual Target-Based
Database Screening
Yongping Pan, Niu Huang, Sam Cho, and Alexander D. MacKerell, Jr.*
Department of Pharmaceutical Sciences, School of Pharmacy, University of Maryland,
Baltimore, Maryland 21201
Received September 10, 2002
Virtual database screening allows for millions of chemical compounds to be computationally selected based
on structural complimentarity to known inhibitors or to a target binding site on a biological macromolecule.
Compound selection in virtual database screening when targeting a biological macromolecule is typically
based on the interaction energy between the chemical compound and the target macromolecule. In the present
study it is shown that this approach is biased toward the selection of high molecular weight compounds due
to the contribution of the compound size to the energy score. To account for molecular weight during energy
based screening, we propose normalization strategies based on the total number of heavy atoms in the
chemical compounds being screened. This approach is computationally efficient and produces molecular
weight distributions of selected compounds that can be selected to be (1) lower than that of the original
database used in the virtual screening, which may be desirable for selection of leadlike compounds or (2)
similar to that of the original database, which may be desirable for the selection of drug-like compounds.
By eliminating the bias in target-based database screening toward higher molecular weight compounds it is
anticipated that the proposed procedure will enhance the success rate of computer-aided drug design.
INTRODUCTION
Computer-based virtual database screening for lead compounds has become one of the integral approaches in the
structure-based drug discovery process.1 Database searching
or docking approaches that select small molecular weight
compounds that are structurally or energetically complementary to a putative binding site on a biological macromolecule have been reasonably successful in the past few
years.2-6 However, the hit rate, or percentage of biologically
active compounds out of those selected via the screening
procedure, is typically 20% or less indicating the need for
further improvements in the selection procedure.6-8
A key element in determining the hit rate of virtual
screening is the energy function used for prediction of the
binding orientation and energy.7,9,10 Currently, compounds
are selected based on a variety of energetic scoring methods.
For example, Chen et al.8 and Enyedy et al.11 identified lead
inhibitors targeting HIV-1 intergrase (IN) and Bcl-2, respectively, based on the empirical energy function implemented
in DOCK.12 A similar energy function with preferences for
hydrogen bonds is implemented in FlexX.13 Other energy
based scoring methods include potentials of mean force
(PMF),14-16 free energy grids17 and many others.2,9 An
extension of these approaches is the consensus-scoring
method,18,19 that combines two or more scoring functions.
This method was suggested to be more robust than using
the individual scoring functions, but the hit rate was
comparable due to the inherent limitation of the energy
functions.20
* Corresponding author phone: (410)706-7442; fax: (410)706-0346;
e-mail: amackere@rx.umaryland.edu. Corresponding author address: 20
N. Pine Street, Baltimore, MD 21201.
One important issue that has not been adequately dealt
with to date in most docking approaches, and that may
adversely influence the hit rate, is the molecular weight
(MW) of the selected molecules, though several studies have
mentioned its impact on compound selection.14,21 It is
accounted for partially in the energy function implemented
in FlexX13 by including the number of rotatable bonds in
the scoring criteria, which was intended to account for the
entropy penalty for constraining the rotatable bonds upon
binding. While such a correction is appropriate it does not
account for the fact that the overall van der Waals (VDW)
interaction energy, a term suggested to be important for
ligand affinity,22 is the sum over all pairs of ligand and target
protein atoms within a specified cutoff distance. Therefore,
energetic selection of compounds that includes a VDW
attractive contribution, or similar term, will favor larger MW
compounds since they have a larger number of atoms
interacting with the target molecule. Thus, there will be a
tendency for the selection of larger molecules even though
they may not necessarily be as structurally complementary
to the target binding site as smaller compounds. Adjustment
of the energy score to account for the size of the molecule
may correct for this problem. In the present manuscript, a
normalization method to adjust the energy score based on
the total number of heavy atoms in a molecule is presented.
Based on this approach the tendency of energy based scoring
methods to bias toward the selection of higher molecular
weight compounds is controlled.
Previous work8 in our laboratory has targeted the Y3binding site on HIV-1 integrase23 via database screening of
the National Cancer Institute (NCI) nonproprietary database.24 Here we chose the same target and two other
databases, the Derwent World Drug Index (WDI)25 and the
10.1021/ci020055f CCC: $25.00 © 2003 American Chemical Society
Published on Web 11/14/2002
268 J. Chem. Inf. Comput. Sci., Vol. 43, No. 1, 2003
CHEMDIV26 database, to test the influence of MW on the
compounds selected using the DOCK energy scoring function12 directly and following normalization. It is shown that
based on the normalization strategy applied, compound
selection could be biased to (1) a distribution that had MWs
reminiscent of those associated with leadlike compounds27
or (2) a distribution of MWs that significantly overlapped
with the entire database MW distribution.
METHODS
Two chemical databases, WDI and CHEMDIV with
approximately 50 000 and 200 000 compounds, respectively,
were used for docking to the Y3 binding site of HIV-1
integrase. Three-dimensional (3D) structures of the compounds in the databases were generated as follows: the
2-dimensional (2D) structural data files (SDF) from the
vendors were first converted to 2D MOL2 format using the
program SDF2MOL2 implemented in DOCK 4.0.1.12 Hydrogens and Gasteiger charges were then calculated and
added to the 2D MOL2 structures followed by 500 steps of
minimization with the POWELL conjugate gradient algorithm using the SYBYL6.728 molecular modeling package.
The protonation state of the ligands was taken directly from
the suppliers.
DOCK 4.0.112 was used to carry out the database screening. Modeling of the binding site and selection of the sphere
set were described previously.8 The anchor-search-first
algorithm was used to initially place the ligands and their
conformational space was searched via the standard torsion
drive. While the added segment was minimized at every
intermediate stage, re-minimize-layer-numbers of 3 and 5
were used for the large database screening (method one) and
for the second, more rigorous search (method two), respectively. Anchor re-minimization during the conformational
search was also used in method two. A dielectric function
of 4r and a cutoff of 10.0 Å were used for generating the
steric and electrostatic environment at the binding site with
the GRID29 module of DOCK. In all cases the docking and
resulting binding orientations were based on the total energy,
such that selection of compounds via the VDW attractive
energy used the docked orientations based on the total
energy. Heavy atom cutoff numbers of 40 and 60 were used
to compare its influence on compound selection. Solvent
accessible surface areas30 of the small molecules were
calculated with CHARMM31 program and the MMFF32 force
field implemented in CHARMM using a 1.4 Å probe radius.
RESULTS AND DISCUSSIONS
In the following, results are presented showing how
database screening using the program DOCK leads to the
selection of compounds with a higher MW distribution than
those in the original database. A normalization procedure is
then proposed and tested that corrects for this bias. Two
databases have been selected for the study. The WDI was
selected as it contains a collection of pharmacologically
active, or druglike, compounds, including therapeutic agents
currently on the market. Since it may be considered undesirable to shift the MW distribution of selected compounds
above that of known pharmacologically active compounds,
inclusion of a database of biologically active compounds is
important. Alternatively, it is necessary to check if the
PAN
ET AL.
Table 1. Mean Molecular Weights and Standard Errors for the
WDI and CHEMDIV Databases and for the 5000 Compounds
Selected Using Different Scoring and Normalization Proceduresa
vdw attractive
normalization
40
60
40
WDI Full Database Mean 359 ( 8
369 ( 12
382 ( 12
389 ( 15
258 ( 24
258 ( 24
284 ( 7
289 ( 6
290 ( 6
320 ( 7
310 ( 5
313 ( 5
339 ( 10
332 ( 10
338 ( 9
359 ( 11
none
N
N2/3
N1/2
N1/3
vdw attractive
normalization
none
N
N2/3
N1/2
N1/3
total energy
40
60
60
403 ( 16
284 ( 7
322 ( 7
343 ( 9
365 ( 12
total energy
40
CHEMDIV Full Database Mean 385 ( 7
431 ( 11
440 ( 13
442 ( 14
305 ( 13
305 ( 12
309 ( 13
346 ( 8
346 ( 7
350 ( 10
369 ( 2
370 ( 2
375 ( 5
393 ( 3
395 ( 3
400 ( 5
60
453 ( 19
309 ( 13
350 ( 10
377 ( 4
404 ( 6
a Mean and standard error values were calculated by separating each
data set into 5 independent data sets (e.g. a set of 5000 compounds
MWs were separated into 5 1000 compound MW sets), the mean
obtained for each individual set and then obtaining the mean and the
standard error over the 5 individual means.36
problem and resulting solution are also applicable to a
chemical database typically used for database searching. This
motivated the selection of CHEMDIV, a diverse collection
of compounds designed for database searching.
Molecular Weight Distributions from Database Screening. The MW distributions of the 5000 selected compounds
based on both the VDW attractive and total energy scores
are presented in Figure 1A,B, for the WDI and CHEMDIV
databases, respectively. A significant observation is that all
the distributions of the docked compounds are shifted toward
the high MW region relative to the full database distributions
in both cases. These results are quantitated in Table 1, where
it can be seen that the mean MW of the 5000 selected
compounds (normalization: none) are larger than that of the
original database in all cases. For example, with the WDI
the mean MW of the entire database is 359, while that from
the scoring based on the VDW attractive term with a 40 nonhydrogen atom cutoff is 369 and increases to 382 with a 60
atom cutoff. Similar increases in the mean MW are seen with
the total energy scoring as well as with the CHEMDIV
database.
The results based on the VDW attractive or total energy
scores are similar since total scores for most molecules were
dominated by the VDW attractive energy (see Supporting
Information Figures S1 and S2). These results show that
there is a bias toward the selection of high MW compounds
when doing target-based database screening, leading to MW
distributions skewed to values larger than those typically
seen for pharmacologically active compounds as judged
by the WDI. As mentioned above, similar upshifting of the
MW of selected compounds has also been reported in
database screening work based on PMF scoring.14 Clearly,
a normalization procedure to correct for the bias toward
higher MW compounds would be desirable, as lower MW
compounds typically have improved absorption and disposition properties.33 In addition, when considering the use of
database screening in drug design, it may be considered
VIRTUAL TARGET-BASED DATABASE SCREENING
Figure 1. Molecular weight distribution for the top 5000 molecules
selected from the (A) WDI and (B) CHEMDIV databases based
on different scoring: VDW attractive energy and cutoff 60 (red)
and total energy and cutoff 60 (green). For comparison, the original
database distributions are included (blue).
desirable to select compounds with lower MW than those
seen in the WDI as they may be more appropriate for lead
compound structural optimization.34
Score Normalization Based on the Number of NonHydrogen Atoms. Empirically, the bias toward the selection
of higher MW compounds in database screening is not
surprising given that additional favorable interactions between a ligand and a receptor are available to larger
compounds. This is especially true for the VDW attractive
term, which is favorable for all atom pairs and leads to
consideration of a normalization procedure based on the
number of non-hydrogen atoms, N, in each ligand. This is
similar to the use of scoring on a per heavy atom basis as
used in the SmoG scoring function.35 Accordingly, the VDW
attractive or total energy for each compound was divided
by the number of non-hydrogen atoms in each respective
compound and the top 5000 compounds selected based on
those normalized scores. Presented in Figure 2 are the
resulting MW distributions for the two databases and for the
different scoring criteria. As is evident, score normalization
based on N leads to a significant downshift in the MW of
the selected compounds as compared to the total databases.
Consistent with this shift are the mean MWs reported in
Table 1 for N normalization. Thus, use of the number of
non-hydrogen atoms for score normalization leads to a
significant overcorrection, leading to a strong bias toward
low MW compounds. However, such an overshift may be
considered desirable in database screening efforts where the
goal is the identification of leadlike versus druglike compounds.34
Alternatively, a normalization procedure may be considered that is based on the surface area of a molecule. In this
model, the increase in the surface area of a molecule as the
number of atoms increases would allow for more favorable
interactions, thereby biasing toward high MW compounds
during screening. However, determination of the solvent
accessible surface area of a molecule is computationally
J. Chem. Inf. Comput. Sci., Vol. 43, No. 1, 2003 269
Figure 2. Molecular weight distribution for the top 5000 molecules
selected using the N normalization scoring method from the (A)
WDI and (B) CHEMDIV databases based on different scoring:
normalized VDW attractive energy and cutoff 60 (red) and
normalized total energy and cutoff 60 (green). For comparison, the
original database distributions are included (blue).
expensive, especially in situations where it must be determined for a million or more molecules, a situation common
in virtual database screening. Instead, one may consider that
N is directly proportional to the volume of a molecule and,
assuming a simple geometry, is also proportional to the radius
cubed, r3. Based on r being proportional N1/3 and the surface
area being proportional to r2, the surface area is proportional
to N2/3. To test this model, the solvent accessible surface
area of 4523 randomly selected molecules from the CHEMDIV database was calculated and is plotted against N2/3 in
Figure 3. As expected a linear relationship is present (R )
0.91). Accordingly, the VDW attractive and total energy
scores were renormalized using N2/3 and the resulting MW
distributions presented in Figure 4. For both WDI and
CHEMDIV, there is still a downshifting of the MW
distribution below those in the entire databases, though the
effect is smaller than that observed with N normalization
(Figure 2). The mean MWs in Table 1 show the values for
the N2/3 selected compounds to be smaller than the mean
values from the entire databases.
Motivated by the shift in MW distributions upon going
from N to N2/3 normalization, tests using N1/2 normalization
were undertaken. Presented in Figure 5 and Table 1 are the
MW distributions and mean MWs, respectively, following
rescoring with normalization by N1/2. Both Figure 5 and
Table 1 show that this normalization brings the MW
distribution of selected compounds into better agreement with
that of the entire databases, although a shift to small MWs
is evident. Therefore, normalization via N1/3 was tested, with
the resulting MW distributions shown in Figure 6 and the
mean values in Table 1. With the WDI, the selected compounds are still shifted to lower MWs, while with CHEMDIV
the MW distribution of the selected compounds is in good
agreement with that of the full database. The larger discrepancy with the WDI appears due to the number of high MW
compounds in the entire database, which increases the mean.
270 J. Chem. Inf. Comput. Sci., Vol. 43, No. 1, 2003
PAN
ET AL.
Figure 3. Two-thirds root of the number of heavy atoms, N2/3, versus the solvent accessibility for 4523 compounds randomly selected
from the CHEMDIV database. The line represents the least-squares fit to the data.
Figure 4. Molecular weight distribution for the top 5000 molecules
selected using the N2/3 normalization scoring method from the (A)
WDI and (B) CHEMDIV databases based on different scoring:
normalized VDW attractive energy and cutoff 60 (red) and
normalized total energy and cutoff 60 (green). For comparison, the
original database distributions are included (blue).
Figure 5. Molecular weight distribution for the top 5000 molecules
selected using the N1/2 normalization scoring method from the (A)
WDI and (B) CHEMDIV databases based on different scoring:
normalized VDW attractive energy and cutoff 60 (red) and
normalized total energy and cutoff 60 (green). For comparison, the
original database distributions are included (blue).
Thus, the empirically motivated use of N1/3 for score
normalization yields a MW distribution of selected compounds that is most similar to that of the original database
being screened. However, the use of N1/3 is difficult to justify
based on physical considerations. It is therefore suggested
that the N1/3 normalization factor represents a collection of
physical phenomena, including the surface area term suggested above. It should be noted that the relationship of
surface area to N1/3 is almost identical to that of N2/3, with
R ) 0.91.
The “rule of five” used for lead identification and
optimization33 lends additional support to the advantage of
the proposed normalization procedure. The decreased MW
distributions may be considered desirable as compounds with
MWs greater than 500 daltons will typically have lower
biological membrane permeability and therefore lower bioavailability.33 This is supported by a previous database
screening study where selected compounds that were biologically active had a lower MW distribution than drugs on
the market.34
VIRTUAL TARGET-BASED DATABASE SCREENING
J. Chem. Inf. Comput. Sci., Vol. 43, No. 1, 2003 271
Figure 7. Probability distributions of the ratio of the buried surface
area upon going from the unbound to bound states for the original
(squares) and normalized (circles) energy scores. Results are based
on the change in the surfaces areas for the bound conformations of
the ligands to IN from the WDI VDW energy database screen.
account. The higher ratio of contacts to surface area may
facilitate the selection of compounds with a higher probability
of being specific for the target molecule.
Figure 6. Molecular weight distribution for the top 5000 molecules
selected using the N1/3 normalization scoring method from the (A)
WDI and (B) CHEMDIV databases based on different scoring:
normalized VDW attractive energy and cutoff 60 (red) and
normalized total energy and cutoff 60 (green). For comparison, the
original database distributions are included (blue).
To further illustrate the influence of N1/3 normalization
on the MW distribution of the selected compounds, the 5000
compounds selected from the WDI database using the
method one screening protocol based on VDW attractive
energy were further screened using method two with 500
compounds selected based on the total energy score. Without
normalization after method one and method two screening,
121 out of 500 compounds had MWs greater than 500
daltons, compared with 38 out of 500 compounds with N1/3
normalization used in both steps. The mean molecular
weights of the distributions were 447 and 367 daltons for
un-normalized and normalized cases, respectively, versus 359
(Table 1) for the entire WDI database. Consistent with the
results presented above, the more rigorous method two
docking protocol was again biased toward the higher MW
region without normalization.
Influence of Normalization on Ligand-Target Complimentarity. A final question was whether the proposed
normalization procedure enhances the complimentarity between the ligand and the target binding site. This was
addressed by analyzing the ratio of the buried surface areas
of the ligands in the bound state relative to that of the
unbound ligand for the unnormalized and N1/2 normalized
scores for the WDI method one screen using VDW attractive
energy and 60 atom cutoff. As is shown in Figure 7, the
normalization procedure shifts the distribution toward compounds with a higher ratio of buried surface area. This
indicates that compounds selected via the normalized score
show a tendency to have a higher percentage of their surface
area buried in the bound state versus the unbound state than
those selected via the un-normalized scores. While certainly
not an unequivocal measure of complimentarity, these results
suggest that the normalization procedure may favor the
selection of compounds with more contacts per unit surface
area with the target binding site when MW is taken into
SUMMARY
The most important issue in virtual database screening for
lead compounds is the ability of the protocol used to generate
a satisfying hit rate. In this work it is shown that use of
interaction energy scoring alone for choosing molecules
favors the selection of high MW compounds. However, by
normalizing the energy score based on the number of heavy
atoms, N, in the ligands the bias toward higher MW
compounds can be eliminated. Moreover, by selection of the
normalization procedure based on the power of N, a target
range for the MW distribution can be selected. For example,
if the goal of a study is the identification of druglike
compounds,34 normalization via N1/3 would be appropriate
as it yields a distribution similar to that of the entire database,
whereas if leadlike molecules are the goal, normalization via
N is appropriate as it yields a MW distribution for the
selected compounds with mean values of 300 daltons or less
(Table 1). It is anticipated that this approach may enhance
the hit-rate of database screening efforts.
In the present study we have applied our normalization to
only a single binding site on HIV IN using the program
DOCK and its energy scoring function.29 However, in work
by Muegge et al.,14 a similar upshift of the MW distribution
was observed using DOCK with a PMF based scoring
function.15 Thus, the proposed approach is anticipated to be
applicable to a variety of target-based database screening
studies, although it is suggested that N to various powers be
tested to identify the value that yields the desired MV
distribution appropriate for each particular project.
ACKNOWLEDGMENT
The authors thank Dr. Holger Gohlke for helpful discussions and financial support from the NIH (GM CA95200),
and the Greenebaum Cancer Center, the Department of
Pharmaceutical Sciences, School of Pharmacy, Baltimore,
and the Computer Aided Drug Design Center, School of
Pharmacy, University of Maryland, Baltimore.
Supporting Information Available: Figures of energy
component distribution for the 5000 compounds selected from
the CHEMDIV and WDI databases. This material is available
free of charge via the Internet at http://pubs.acs.org.
272 J. Chem. Inf. Comput. Sci., Vol. 43, No. 1, 2003
REFERENCES AND NOTES
(1) Walters, W. P.; Stahl, M. T.; Murcko, M. A. Virtual screening-an
overview. Drug DiscoVery Today 1998, 3, 160-178.
(2) Gane, P. J.; Dean, P. M. Recent advances in structure-based rational
drug design. Curr. Opin. Struct. Biol. 2000, 10, 401-404.
(3) Burkhard, P.; Hommel, U.; Sanner, M.; Walkinshaw, M. D. The
discovery of steroids and other novel FKBP inhibitors using a
molecular docking program. J. Mol. Biol. 1999, 287, 853-858.
(4) Godden, J. W.; Stahura, F.; Bajorath, J. Evaluation of docking strategies
for virtual screening of compound databases: camp-dependent serine/
threonine kinase as an example. J. Mol. Graphics Mod. 1999, 16, 139143.
(5) Filikov, A. V.; Monan, V.; Vickers, T. A.; Griffey, R. H.; Cook, P.
D.; Abagyan, R. A.; James, T. L. Identification of ligands for RNA
targets via structure-based virtual screening: HIV-1 TAR. J. Comput.
Aided Mol. Des. 2000, 14, 593-610.
(6) Perola, E.; Xu, K.; Kollmeyer, T. M.; Kaufmann, S. H.; Prendergast,
F. G.; Pang, Y.-P. Successful virtual screening of a chemical database
for farnesyltransferase inhibitor leads. J. Med. Chem. 2000, 43, 401408.
(7) Bissantz, C., Folkers, G.; Rognan, D. Protein-Based Virtual Screening
of Chemical Databases. 1. Evaluation of Different Docking/Scoring
Combinations. J. Med. Chem. 2000, 43, 4759-4767.
(8) Chen, I.-J.; Neamati, N.; Nicklaus, M. C.; Or, A.; Anderson, L.; Barchi,
J. J., Jr.; Kelley, J. A.; Pommier, Y.; MacKerell, A. D., Jr. Identification
of HIV-1 Integrase Inhibitors via Three-Dimensional Database Searching Using ASV and HIV-1 Integrases as Targets. Bioorg., Med. Chem.
2000, 8, 2385-2398.
(9) Stahl, M.; Rarey, M. Detailed analysis of scoring functions for virtual
screening. J. Med. Chem. 2001, 44, 1035-1042.
(10) Pearlman, D. A.; Charifson, P. S. Are free energy calculations useful
in practice? A comparison with rapid scoring functions for the p38
MAP kinase protein system. J. Med. Chem. 2001, 44, 3417-3423.
(11) Enyedy, I. J.; Ling, Y.; Nacro, K.; Tomita, Y.; Wu, X.; Cao, Y.; Guo,
R.; Li, B.; Zhu, X.; Huang, Y.; Long, Y.-Q.; Roller, P. P.; Yang, D.;
Wang, S. Discovery of Small-Molecule Inhibitors of Bcl-2 through
Structure-Based Computer Screening. J. Med. Chem. 2001, 44, 43134324.
(12) Makino, S.; Kuntz, I. D. Automated Flexible Ligand Docking Method
and Its Application for Database Search. J. Comput. Chem. 1997, 18,
1812-1825.
(13) Rarey, M.; Kramer, B.; Lengauer, T.; Klebe, G. A fast flexible docking
method using an incremental construction algorithm. J. Mol. Biol.
1996, 261, 470-489.
(14) Muegge, I.; Martin, Y. C.; Hajduk, P. J.; Fesik, S. W. Evaluation of
PMF Scoring in Docking Weak Lignads to the FK506 Binding Protein.
J. Med. Chem. 1999, 42, 2498-2503.
(15) Muegge, I.; Martin, Y. C. A General and Fast Scoring Function for
Protein-Ligand Interactions: A simplified Potential Approach. J. Med.
Chem. 1999, 42, 791-804.
(16) Gohlke, H.; Hendlich, M.; Klebe, G. Knowledge-based Scoring
Function to Predict Protein-Ligand Interactions. J. Mol. Biol. 2000,
295, 337-356.
(17) Pearlman, D. A.; Charifson, P. S. Improved Scoring of Ligand-Protein
Interactions Using OWFEG Free Energy Grids. J. Med. Chem. 2001,
44, 502-511.
PAN
ET AL.
(18) Charifson, P. S.; Corkery, J. J.; Murcko, M. A.; Walters, W. P.
Consensus Scoring: A Method for Obtaining Improved Hit Rates from
Docking Databases of Three-Dimensional Structures into Proteins. J.
Med. Chem. 1999, 42, 5100-5109.
(19) Gohlke, H. and Klebe, G. Statistical potentials and scoring functions
applied to protein-ligand binding, Curr. Opin. Struct. Biol. 2001, 11,
231-235.
(20) Wang, R.; Wang, S. How Does Consensus Scoring Work for Virtual
Library Screening? An Idealized Computer Experiment. J. Chem. Inf.
Comput. Sci. 2001, 41, 1422-1426.
(21) Pegg, S. C.-H.; Haresco, J. J.; Kuntz, I. D. A genetic algorithm for
structure-based de novo design. J. Comput.-Aided Mol. Des. 2001,
15, 911-933.
(22) Kuntz, I. D.; Chen, K.; Sharp, K.; Kollman, P. A. The maximal affinity
of ligands. Proc. Natl. Acad. Sci. U.S.A. 1999, 96, 9997-10002.
(23) Craigie, R. HIV Integrase, a brief overview from chemistry to
therapeutics. J. Biol. Chem. 2001, 276, 23213-23216.
(24) Milne, G. W. A.; Nicklaus, M. C.; Driscoll, J. S.; Wang, S.; Zaharevitz,
D. National Cancer Institute Drug Information System 3D Database.
J. Chem. Inf. Comput. Sci. 1994, 34, 1219-1224.
(25) WDI: World Drug Index; Version 4/99, Derwent Information, 1999
(www.derwent.com).
(26) See http://www.chemdiv.com.
(27) Teague, S. J.; Davis, A. M.; Leeson, P. D.; Oprea, T. The Design of
Leadlike Combinatorial Libraries. Angew. Chem., Int. Ed. Engl. 1999,
38, 3743-3748.
(28) SYBYL, version 6.7; Tripos Associates: St. Louis MO, 2000.
(29) Meng, E. C.; Shoichet, B. K.; Kuntz, I. D. Automatic docking with
grid-based energy evaluation. J. Comput. Chem. 1992, 13, 505524.
(30) Lee, B.; Richards, F. M. The interpretation of protein structures: estimation of static accessibility. J. Mol. Biol. 1971, 55, 379400.
(31) Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.;
Swaminathan, S.; Karplus, M. CHARMM: A program for macromolecular energy, minimization, and dynamics calculations. J. Comput.
Chem. 1983, 2, 187-217.
(32) Halgren, T. A. Merck molecular force field: I. Basis, form, scope,
parametrization and performance of MMFF. J. Comput. Chem. 1996,
17, 490-519.
(33) Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeney, P. J.
Experimental and computational approaches to estimate solubility and
permeability in drug discovery and development settings. AdV. Drug
DeliVery ReV. 2001, 46, 3-26.
(34) Oprea, T. I., Davis, A.; Teague, S. J.; Leeson, P. D. Is there a difference
between leads and drugs? A historical perspective. J. Chem. Inf.
Comput. Sci. 2001, 41, 1380-1315.
(35) DeWitte, R. S.; Shaknovitch, E. I. SMoG: de Novo Design Method
Based on Simple, Fast and Accurate Free Energy Estimates. 1.
Methodology and Supporting Evidence. J. Am. Chem. Soc. 1996, 118,
11733-11744.
(36) Loncharich, R. J.; Brooks, B. R.; Pastor, R. W. Langevin Dynamics
of Peptides: The Frictional Dependence of Isomerization Rates of
N-Acetylalanyl-N′-Methylamide. Biopolymers 1992, 32, 523-535.
CI020055F
Download PDF
Similar pages