Chem-Bio Informatics Journal, Vol.10, pp.

74-86(2010)

De novo Based Ligand generation and Docking studies of

PPARδ Agonists: Correlations between Predicted Biological
activity vs. Biopharmaceutical Descriptors

Vasudeva Rao Avupati*, Purna Nagasree Kurre, Santoshi Rupa Bagadi,

Muralikrishna Kumar Muthyala, and Rajendra Prasad Yejella

University College of Pharmaceutical Sciences, Pharmaceutical Chemistry Division, Andhra University,

Visakhapatnam-530003, Andhra Pradesh, India. address
*E-mail: vasudevaraoavupati@gmail.com

(Received January 24, 2010; accepted May 15, 2010; published online June 3, 2010)

Abstract
Molecular docking was performed on a series of bisaryl substituted thiazoles and oxazoles
as PPARδ agonists. The docking technique was applied to dock a set of representative
compounds within the active site region of 3D5F using Molegro Virtual Docker v 4.0.0. For
these compounds, the correlation between binding free energy (kcal/mol) and log (1/EC50)
values produces a good correlation coefficient (r2 = 0.719). The docking simulation clearly
predicted the binding mode that is nearly similar to the crystallographic binding mode within
0.91Å RMSD. Based on the validations and interactions made by Ar1 and Ar2 substituents,
ligand design was initiated considering simple combinations. For the designed compounds
biopharmaceutical properties e.g. Lipophilicity (logP), Solubility (logS), Ionization constant
(pKa), Distribution coefficient (logD) are predicted computationally using ACD/ChemSketch
v 12.0. The hydrogen bond interactions are examined and bivariate statistical correlation
between predicted biological activity (log (1/EC50) and biopharmaceutical properties are
considered for evaluation. Ligand 11 (cC) thus, showed high binding energy (-206.73
kcal/mol) against PPARδ. The results avail to understand the type of interactions that occur
between designed ligands with PPARδ binding site region and explain the importance of Ar1
and Ar2 substitutions on derivatives of bisaryl substituted thiazoles and oxazoles.

Key Words: Moldock Scores, biopharmaceutical properties, bivariate statistical correlation, PPARδ delta, ligand
design

Area of Interest: Computer Aided Drug Design

74 Copyright 2010 Chem-Bio Informatics Society http://www.cbi.or.jp

Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

1. Introduction

Peroxisome proliferator activated receptors (PPARs) are important members of the nuclear
hormone receptor superfamily. These receptors are ligand activated transcription factors known to
play a key role in the catabolism and storage of dietary fats [1]. Three PPAR isotypes: PPARα
(NR1C1), PPARδ (also called β {NR1C2}), PPARγ (NR1C3) have been identified so far. Once
activated by their respective ligand, PPARs control transcriptional rate of a large panel of genes
implicated in various physiological functions, including lipid and glucose homeostasis,
inflammation, cell proliferation and differentiation. The various PPAR isotypes have different
physiological roles [2]. PPARδ is primarily expressed in adipose tissue, where it acts as a master
regulator of adipocyte formation [3].
Like other PPARs, PPARδ protein molecular structure consists of 5 regions : an N-terminal
region (A/B), a DNA binding domain (C), a flexible hinge region (D), ligand binding domain (E),
and 2-C-terminal region (F). X-ray crystallographic study revealed that PPARδ has an exceptionally
large ligand binding pocket [4], [5]. The ligand binding pocket of PPARδ receptor is considerably
larger than other nuclear receptors displaying a total volume of ~1300 Å [6]. The pocket forms a
“Y” shape comprised of three arms approximately 12 Å in length [5][4]. The Y-shaped molecules
are potent and selective PPARδ agonists and the chirality at the Y intersection is pivotal to PPARδ
agonist activity [7].
Recently, several synthetic ligands have been reported to selectively activate PPARδ include
3,4,5-trisubstituted isoxazoles [8], 1,3,5-trisubstituted aryls [9], benzothiophenes, benzofuran and
indole based compounds [10], anthranilic acid GW9371 [11], phenylpropanoic acid derivatives
bearing 6-substituted benzothiazoles [12], para-alkyl thio phenoxy aceticacids [7]. The phenoxy
acetic acid derivatives GW501516 and GW0742 are the highly selective PPARδ ligands with
nanomolar affinity and 1000-fold selectivity over other isotypes PPARα and PPARγ [13]. Novel
bisaryl substituted thiazoles and oxazoles are highly potent and selective PPARδ agonists [14].
The other PPARδ agonists L796449, L165461 [15], KD3010 and MBX-8025 are currently in
clinical development. A selective antagonist for PPARδ, GSK0660 has recently reported to exhibit
inverse agonist activity and competes with agonist in cellular context [16]. However, because of
low degree of selectivity and specificity against PPARδ and to design more potent and highly
selective PPARδ agonists, modulation of these compounds is required. Many of the compounds
identified by High Throughput Screening (HTS) have poor biopharmaceutical properties [17], and
many thus fail during preclinical development or need complex delivery systems in order to be
effective. Thus there is a need for rapid and efficient computational methods capable of
differentiating compounds with acceptable biopharmaceutical properties, e.g. solubility,
lipophilicity, ionization constant etc at an early stage in the drug discovery process. In the present
study, structure based drug design stratagies were employed on bisaryl substituted thiazoles and
oxazoles. Through in silico docking procedures different modes of interactions exhibited by these
synthesized ligands will be recognized and further examined for their predicted biological activity
vs. biopharmaceutical properties.

75
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

2. Materials and Methods Section

2.1 Data set

A set of novel PPARδ agonists listed in Table 1, were taken from the biological data reported by
Epple et al [14].

76
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

2.2 Computational Methodology

Software Molegro Virtual Docker (MVD) v 4.0.0 along with graphical user interface, MVD
tools was utilized to generate grid, calculate dock score and evaluate conformers. The structures of
compounds were drawn using Chemdraw ultra v 10.0 (Cambridge software), copied to Chem3D
ultra v 10.0 to create a 3D model and, finally subjected to energy minimization using molecular
mechanics (MM2). The minimization was executed until the root mean square gradient value
reached a value smaller than 0.001kcal/mol. Such energy minimized structures are considered for
docking and corresponding pdb files were prepared using Chem3D ultra v 10.0 integral option
(save as /Protein Data Bank (pdb)). The selection of protein for docking studies is based upon
several factors i.e. structure should be determined by x-ray diffraction, and resolution should be
between 2.0-2.5Å, it should contain a co-crystallized ligand. Among 12 entries of PPARδ proteins
deposited in Protein Data Bank (PDB), 3D5F was selected for docking analysis based on
ramachandran plot statistics as it showed 414 amino acid residues in most favored regions, 43 in
additionally allowed regions and none of the residues in disallowed regions.

2.3 Validation of the docking method

Software validation was performed in MVD using PDB protein 3D5F. The x-ray crystal
structure of PPARδ complex with L41501 was recovered from PDB. The bio active co-crystallized
bound ligand {4-[3-(4-acetyl-3-hydroxy-2-propylphenoxy) propoxy] phenoxy} acetic acid
(L41501) was docked within the active site region formed by Trp 264, Val 341, Arg 284, Leu 330,
Lys 367, Ileu 364, Phe 282, Met 453, Thr 289, Leu 469, His 323, Cys 285, His 449, Thr 288 and
Tyr 473 residues, respectively. The RMSD of all atoms between the two conformations is 0.91Å
indicating that the parameters for docking simulation are good in reproducing X-ray crystal
structure. Similar software validations were performed for the rest of PDB proteins i.e. 3GWX,
2B50, 1Y0S, 2ZNP, 2ZNQ, 2J14, 3ET2, 3GZ9, 3DY6, 2BAW and 2AWH respectively. The
RMSD values were ranging from 0.91-1.75Å as shown in Table 2. Another validation was carried
out based on the important interactions made by the bound ligand with active site residues of all
PPARδ proteins from PDB. In order to perform the task, the H-bond interactions formed by various
bound co-crystallized ligands of PPARδ protein structures such as 3D5F, 3GWX, 2B50, 1Y0S,
2ZNP, 2ZNQ, 2J14, 3ET2, 3GZ9, 3DY6, 2BAW and 2AWH were collected from Ligplot
interactions, deposited in PDB summary (PDB sum, http//www.ebi.ac.uk) database and listed in
Table 2.

77
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

2.4 Molecular Docking

In this article, we introduce a docking algorithm called MolDock. MolDock is based on a new
hybrid search algorithm, called guided differential evolution. The guided differential evolution
algorithm combines the differential evolution optimization technique with a cavity prediction
algorithm. Differential evolution (DE) was introduced by Storn and Price [18] in 1995 and has
previously been successfully applied to molecular docking [19]. The use of predicted cavities
during the search process, allows for a fast and accurate identification of potential binding modes
(poses). The docking scoring function of MolDock that we use is based on a piecewise linear
potential (PLP) introduced by Gehlhaar et al [20] [21] and further extended in GEMDOCK by Yang
et al [22]. We used MVD because it showed higher docking accuracy than other stages of the
docking products (MVD: 87%, Glide: 82%, Surflex: 75%, FlexX: 58%) in the market [23].
Molecular docking technique was employed to dock the selected PPARδ agonists listed in (Table 1)
against PPARδ receptor 3D5F using MVD to locate the interaction between various compounds and
PPARδ. MVD requires the receptor and ligand coordinates in either Mol2 or PDB format. Non
polar hydrogen atoms were removed from the receptor file and their partial charges were added to
the corresponding carbon atoms. Molecular docking was performed using MolDock docking engine
of Molegro software. The binding site was defined as a spherical region which encompasses all
protein atoms within 15.0 Å of bound crystallographic ligand atom (dimensions X (38.52 Å), Y
(31.61 Å), Z (42.08 Å) axes, respectively). Default settings were used for all the calculations.
Docking was performed using a grid resolution of 0.3 Å and for each of the 10 independent runs; a
maximum number of 1500 iterations were executed on a single population of 50 individuals.

2.5 Ligand Design

Novel PPARδ agonist structures were designed based on the compounds selected for docking
experiments. Keeping in view on the chemical groups attached to the structures given in (Table 1),
ligand design was done in such a way that simple modification at Ar1 and Ar2 positions should
retain the properties associated with basic scaffold. Also various Ar1 and Ar2 substituents were
designed by considering the shape of the agonist as well as the residues and shape of the active site
region of 3D5F as shown in Table 3. The designed 16 ligands with various substituents are
presented in (Table 4). Certain chemical rules are utilized to prevent unreasonable structures during
molecular design. For instance, structures that include heteroatoms bonded to each other (e.g. O-O,
N-N and N-O etc) and eliminating too many heteroatoms bonded to the same carbon atom. Also,
certain fragments attached to an aromatic ring possess toxicity. Therefore, in order to recognize
such compounds, a forbidden substructure library from the work of Renxiao Wang et al [24] was
used as a filter.

78
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

79
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

3. Results and Discussion

Docking simulations with 3D5F bound ligand L41501 resulted in a Moldock score of -188.27
kcal/mol and a RMSD value of 0.91Å showed major hydrogen bond interactions with Thr 288, His
323, His 449 and Tyr 473 residues, respectively. The predicted binding conformation of L41501
superimposed with the X-ray crystallographic orientation is shown in Figure 1. The superimposed
binding orientations of docked conformer of 3D5F cocrystallized ligand L41501 with designed
ligand 11(cC) with active site residue interactions were shown in Figure 2 as follows.

3.1 Hydrogen bonding interactions

Docking studies on experimental compounds (Table 1) showed that most of the PPARδ agonists
are involved in hydrogen bonding with residues Tyr 473, Thr 288, Thr 292 and Cys 285 in the
binding site region of 3D5F. Therefore, although other H-bond interactions exist, these hydrogen
bonds are relevant for the binding activities of novel bisaryl substituted thiazoles and oxazoles to be
highly selective and potent PPARδ agonists. Moreover, from the data given in (Table 2), it appears
that the residues Thr 288, Thr 289, His 323, Tyr 473, Lys 319 and Glu 471 represent important
residues for binding diverse range of agonists. Therefore, the important residues that participate in
H-bond interactions as listed above were recognized by our studies on experimental compounds
(Table 1). Then to explore the regions occupied by various substitutions at Ar1 and Ar2 positions in
particular, de novo design of sixteen compounds (Table 3) was carried out. In order to design excess
number of compounds, various possible side chains at Ar1 and Ar2 positions are considered as
shown in Table 3. For combinatorial design, Ar1 substituents are designed with small letters a, b, c,
d, where as Ar2 substituents are referred with capital letters A, B, C and D respectively. From these
eight different Ar1 and Ar2 substitutions, sixteen ligands were generated using simple combinations
such as ‘Ar1 a’ combines with each of Ar2 and ‘Ar1 b’ with each of Ar2 and so on displayed in Table

80
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

3. Further docking procedure using default parameters in these designed ligands resulted in
Moldock scores down to minimum of -206.73 kcal/mol, much lower than the active compound 38f
(-189.11 kcal/mol) and the interacting residues are given in Table 4. The majority of H-bond
interactions formed with residues Thr2 88, Cys 285, Tyr 473 and Ala 342 respectively. From a set
of combinations resulting ‘Ar1’ and ‘Ar2’ substitutions the lowest binding energy was reported in
designed ligand ‘11 cC’ (Table 4) with Moldock Score -206.73 kcal/mol showed 7 H-bond
interactions, shown in Figure 2. Therefore, the in silico approach appears to be useful in predicting
key interacting ligand binding residues. This helps in understanding the type of interactions that
occur between designed ligands with PPARδ binding site region and explain the importance of Ar1
and Ar2 substitutions on novel bisaryl substituted thiazoles and oxazoles. Hence interaction with
Tyr 473, Thr 289, Thr 288, Ala 342, His 280, Glu 259 and Val 281 are significant towards binding
bisaryl substituted thiazoles and oxazoles series of PPARδ agonists.
To further strengthen the approach employed in new ligand generation we examined them
computationally. The EC50 values of newly designed compounds were predicted based on the
correlation between log (1/EC50) (EC50 in µM were converted to negative logarithmic values in
order to assure the linear distribution of data) and the moldock scores (compounds 1 to 19) shown
in Table 1 and established by linear regression technique, according to the regression equation
(Eq)-1, given below.

Observed log (1/EC50) = -0.450364 x (Moldock Score) - 82.9491----- (Eq-1)

In this study we analyzed binding pocket of 3D5F using multiple receptor conformation (MRC)
docking method [26]. We found flexibility in binding pocket residues of 3D5F. The flexibility scale
of amino acids has the following order: Glu > Thr, Val, Tyr, His, Ala > Cys [27]. Hence we resolve
that the seven hydrogen bond interactions shown between conformation of 3D5F and docked
conformer of designed ligand ‘11 cC’ are flexible.

3.2 Calculated ACD/ChemSketch v 12.0 Based Biopharmaceutical Descriptors

The calculations of molecular physical properties by ACD/ChemSketch v 12.0 are based on the
assumption that the properties can be estimated using additive atomic and group increments. LogP
and logD both describe the same physical property lipophilicity. LogD is the appropriate descriptor
for lipophilicity of ionizable compounds because it accounts for the pH dependence of a molecule
in aqueous solution. LogP describes lipophilicity for neutral compounds. Intrinsic solubility logS
affect pH-dependent solubility profile of the compounds. pKa is one of the most challenging tools
to predict multiple ionization systems, tautomerism, charge transfer in conjugated systems, and
multiple ionization centers having a complex affect on the ionization of a particular group.
The ACD/ChemSketch v 12.0 [25] in built ACD/logP program predicts lipophilicity (logP)
based on well characterized logP contributions of separate atoms, structural fragments, and
intramolecular interaction between different fragments these contributions have been derived from
a database of over 6000 structures for which one or more experimental logP values have been
reported in the literature. The most common used solvent system is octanol-1-ol/water. For
ionizable solutes; the compound may exists as variety of different species in each phase at any
given pH. Since ACD/LogD values are appropriate descriptor for the pH dependent differential
solubility of all species in the octanol/water system. ACD/ChemSketch v 12.0 in built ACD/LogD
program predicts values at pH-7.4 based on a structure fragment approach. ACD/ChemSketch v
12.0 in built ACD/Solubility DB program predicts the intrinsic solubility with reference to the
internal database contains over 6000 records, and each record in its turn contains a structure, one or

81
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

more reference fields (i.e., solubility value and reference to the source per field). ACD/ChemSketch
v 12.0 in built ACD/pKa DB program predicts ionization of a particular group (acidic and basic)
made by using methods such as Linear free energy relationships (LFER) based on Hammett and
Taft equations. We utilized ACD/ChemSketch v 12.0 to predict partition coefficient (logP),
distribution coefficient (logD) at pH 7.4, intrinsic solubility (logS), acidic ionization constant (pKa)
of sixteen designed ligands by using default settings (Table 5). The molecular structures were
imported from the ACD/dictionary or drawn using ACD/ChemSketch.

82
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

3.3 Statistical analysis

Bivariate statistical analyses were performed with Molegro Data Modeller v 2.0.0 using the
default settings. Correlation between Moldock Scores and experimental activity log (1/EC50) values
were determined for the data set (Table 1). The statistical relationships between the Moldock Scores
(kcal/mol) and the individual biopharmaceutical descriptors (logP, logD, logS) of the sixteen
designed ligands (Table 5) were determined by using bivariate statistical analysis. Results are
computed and 2D plotted graphs are generated using MDM v 2.0.0 (Figure 3). Statistical results are
shown in following Table 6.

3.3.1 Correlating Moldock Scores vs. log (1/EC50) values of experimental compounds

The computationally predicted Moldock Scores of the experimental compounds (Table 1) used
in our analyses were correlated to their observed biological activity log (1/EC50) values. The
correlation between these two parameters was found to be reliable and produces a good correlation
coefficient (r2 = 0.719) shown in Figure3A. Following is the correlation equation Eq-1.

Observed log (1/EC50) = -0.450364 x (Moldock Score) - 82.9491----- (Eq-1)

3.3.2 Correlating log (1/EC50) vs. ACD/logP, ACD/logD and ACD/logS values of sixteen
designed ligands

The computationally predicted log (1/EC50) values of the sixteen designed ligands (Table 5)
used in our analyses were correlated to their predicted ACD/log P values (Table 5). The correlation
between these two parameters was found to be positive, moderately linear with correlation
coefficient (r2 = 0.515) shown in Figure3B. Following is the correlation Eq-2.

Predicted log (1/EC50) = 1.74417 x (ACD/logP) - 2.60699----- (Eq-2)

83
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

The computationally predicted log (1/EC50) values of the sixteen designed ligands (Table 5)
used in our analyses were correlated to their predicted ACD/log D values at pH 7.4 (Table 5). The
correlation between these two parameters was found to be positive, moderately linear with
correlation coefficient (r2 = 0.532) shown in Figure3C. Following is the correlation Eq-3.

Predicted log (1/EC50) = 1.97143 x (ACD/logD) + 3.12159----- (Eq-3)

The computationally predicted log (1/EC50) values of the sixteen designed ligands (Table 5)
used in our analyses were correlated to their predicted ACD/log S intrinsic solubility values. The
correlation between these two parameters was found to be positive, moderately linear with
correlation coefficient (r2 = 0.565) shown in Figure3D. Following is the correlation Eq-4.

Predicted log (1/EC50) = -2.68571 x (ACD/logS) - 10.4604----- (Eq-4)

4. Conclusions

In this work novel bisaryl substituted thiazoles and oxazoles were investigated using Molegro
Virtual Docker v 4.0.0 software to outline the structural requirements on ligands to discover and

84
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

design the most effective compound as potent PPARδ agonist. Molecular docking studies of
experimentally identified compounds showed an excellent correlation (r2 = 0.719) between binding
free energy (kcal/mol) and log (1/EC50) values against PPARδ. We studied the binding modes
exhibited by various designed ligands illustrate the importance of specific residues forming
H-bonds within the active site region of 3D5F. From our study it is clear that the efficiency of
binding of agonists towards PPARδ would certainly get enhanced when H-bonds are favoured with
Tyr 473, Thr 289, Thr 288, Ala 342, His 280, Glu 259 and Val 281 residues respectively. In the
modern scenario of drug design in addition to designing ligands against any target, necessary care
has to be taken in addressing the importance of biopharmaceutical properties to understand the
behavior of compounds in the real environment under specified conditions. Therefore, we utilized
ACD/ChemSketch v 12.0 based molecular physical properties such as ACD/logP, ACD/logD at pH
7.4, ACD/logS (intrinsic solubility) and ACD/pKa (acidic) are calculated and correlated against
predicted log (1/EC50) values of the sixteen designed ligands. Bivariate statistics applied between
the combinations of predicted log (1/EC50) vs. ACD/logP (r2 = 0.515), predicted log (1/EC50) vs.
ACD/logD at pH 7.4 (r2 =0.532), predicted log (1/EC50) vs. ACD/log S (intrinsic solubility) (r2 =
0.565). A positive moderate linear correlation is well established between predicted log (1/EC50)
and biopharmaceutical properties. The results of this study indicate efficient computational tools are
capable of differentiating potential ligands, even though the pharmacokinetic profile has to be
improved. The utilization of computational tools in the discovery of novel agonists to PPARδ can
be used to save time and reduce the bench work of a chemist.

References

[1] J. Berger, et al., The mechanisms of action of PPARs, Annu. Rev. Med., 3, 409-35(2002).
[2] L. Serge, et al., Roles of PPAR delta in lipid absorption and metabolism: a new target for the
treatment of type 2 diabetes, Biochimica et Biophysica Acta, 1740, 313-317(2005).
[3] E.D. Rosen, et al., Transcriptional regulation of adipogenesis, B. M. Gene Dev., 14,
1293-307(2000).
[4] S.A. Fyffe, et al., Recombinant human PPAR-beta/delta ligand-binding domain is locked in an
activated conformation by endogenous fatty acids, Journal of Molecular Biology, 356(4),
1005-1013(2006).
[5] H.E. Xu, et al., Molecular recognition of fatty acids by peroxisome proliferator-activated
receptors, Molecular Cell, 3(3), 397-403(1999).
[6] R.T. Nolte, et al., Ligand binding and co-activator assembly of the peroxisome
proliferatoractivated receptor-gamma, Nature, 395(6698), 137-143(1998).
[7] R. Zhang, et al., Discovery of para-alkylthiophenoxyacetic acids as a novel series of potentand
selective PPARdelta agonists, Bioorg. Med. Chem. Lett., 17, 3855-9(2007).
[8] R. Epple, et al., 3,4,5-Trisubstituted isoxazoles as novel PPARdelta agonists: Part 1, Bioorg.
Med. Chem. Lett., 16, 4376-4380(2006).
[9] R. Epple, et al., 1,3,5-Trisubstituted aryls as highly selective PPARdelta agonists, Bioorg. Med.
Chem. Lett., 16, 2969-297(2006).
[10] G.F. Filzen, et al., Synthesis and SAR of selective benzothiophene, benzofuran, and
indole-based peroxisome proliferator-activated receptor delta agonists, Bioorg. Med. Chem.
Lett., 17, 3630-363(2007).

85
Chem-Bio Informatics Journal, Vol.10, pp.74-86(2010)

[11] B.G. Shearer, et al., Discovery of a novel class of PPARδ partial agonists, Bioorg. Med. Chem.
Lett., 18, 5018-5022(2008).
[12] H. Fujieda, et al., Phenylpropanoic acid derivatives bearing a benzothiazole ring as
PPARδ-selective agonis, Bioorg. Med. Chem. Lett., 17, 4351-4357(2007).
[13] M.L. Sznaidman, et al., Novel selective small molecule agonists for peroxisomeproliferator-activated
receptor delta (PPARdelta)--synthesis and biological activity, Bioorg. Med. Chem. Lett., 13(9),
1517–1521(2003).
[14] R. Epple, et al., Novel Bisaryl Substituted Thiazoles and Oxazoles as Highly Potent and
Selective Peroxisome Proliferator-Activated Receptor δ Agonists, J. Med. Chem., 53(1),
77–105(2010).
[15] J. Berger, et al., Novel peroxisome proliferator-activated receptor (PPAR) gamma and
PPARdelta ligands produce distinct biological effects, J. Biol. Chem., 274(10),
6718-6725(1999).
[16] B.G. Shearer, et al., Identification and characterization of a selective peroxisome
proliferator-activated receptor beta/delta (NR1C2) antagonist, Molecular Endocrinology,
22(2), 523-529(2008).
[17] C.A. Lipinski, et al., Experimental and computational approaches to estimate solubility
andpermeability in drug discovery and development settings, Adv. Drug Deliv. Rev., 23,
3-25(1997).
[18] R. Storn, et al., Differential Evolution - A Simple and Efficient Adaptive Scheme for Global
Optimization over Continuous Spaces, Technical Report, International Computer Science
Institute, Berkley, CA(1995).
[19] R.Thomsen, Flexible Ligand Docking Using Differential Evolution, Proceedings of the 2003
Congress on Evolutionary Computation, 4, 2354-2361(2003).
[20] D.K. Gehlhaar, et al., Docking Conformationally Flexible Small Molecules into a Protein
Binding Site through Evolutionary Programming. Proceedings of the Fourth International
Conference on Evolutionary Programming, 615-627(1995).
[21] K. Gehlhaar, et al., Fully Automated and Rapid Flexible Docking of Inhibitors Covalently
Bound to Serine Proteases. Proceedings of the Seventh International Conference on
Evolutionary Programming, 449-461(1998).
[22] J.M. Yang, et al., GEMDOCK: A generic evolutionary method for molecular docking,
Proteins, 55(2), 288-304(2004).
[23] R. Thomsen, et al., MolDock: A New Technique for High-Accuracy Molecular Docking, J.
Med. Chem., 49(11), 3315 (2006).
[24] R. Wang, et al., LigBuilder: A Multi-Purpose Program for Structure-Based Drug Design, J.
Mol. Model, 6, 498-516(2000).
[25] ACD/ChemSketch 12.0, Advanced Chemistry Development 33 Richmond St.West, Suite 605,
Toronto, ON MSH 2L3, Canada. http: / / www.acdlabs.com
[26] M. Totrov et al., Flexible ligand docking to multiple receptor conformations: a practical
alternative, Curr. Opin. Struct. Biol., 18(2), 178–184(2008).
[27] Najmanovich, R., et al., Side-chain flexibility in proteins upon ligand binding, Proteins, 39(3),
261-8(2000).

86