1
Traditional Chinese Medicine as Dual
Guardians against Hypertension and Cancer?
Weng Ieong Tou1, 2, Calvin Yu-Chian Chen2, 3, 4, 5, 6*
1School of Medicine, China Medical University, Taichung, 40402, Taiwan
2Laboratory of Computational and Systems Biology, China Medical University,
Taichung, 40402, Taiwan.
3Department of Medical Research, China Medical University Hospital, Taichung,
40402, Taiwan.
4Department of Biotechnology, Asia University, Taichung, 41354, Taiwan.
5Department of Biomedical Informatics, Asia University, Taichung, 41354, Taiwan.
6China Medical University Beigang Hospital, Yunlin,65152, Taiwan.
*Corresponding author. Tel.: +886-4-2205-3366 ext. 4124
E-mail address: ycc929@MIT.EDU (C.Y.-C. Chen)
2
Abstract
This study utilizes the comprehensive traditional Chinese medicine database TCM Database@Taiwan (http://tcm.cmu.edu.tw/) in conjunction with structure-based and ligand-based drug design to identify multi-function Src inhibitors. The three potential TCM candidates identified as having suitable docking conformations and
bioactivity profiles were angeliferulate,
(3R)-2'-hydroxy-3',4'-dimethoxyisoflavan-7-O-beta-D-glucoside(HMID) and 3-[2',6-dihydroxy-5'-(2-propenyl)[1,1'-biphenyl]3-yl]-(E)-2-propenoic acid(3PA).
Molecular dynamics (MD) simulation demonstrated that the TCM candidates have more stable interactions with the cleft and in complex with Src kinase compared to Saracatinib. Angeliferulate and HMID, both originated from Angelica sinensis, not only interact with Lys298 and amino acids from different loops in the cleft, but with Asp407 located on the activation loop. These interactions are important to reduce the opening of the activation loop due to phosphorylation, hence stabilize the Src kinase cleft structure and inhibit activation. The TCM candidates also exhibited high affinity to other cancer-related target proteins (EGFR, HER2, HSP90). Our observations suggest that the TCM candidates might have multi-targeting effects in hypertension and cancer.
Keywords: cancer; Src; traditional Chinese medicine (TCM); docking; QSAR;
molecular dynamics (MD)
3
Introduction
1
Src-family protein kinases are proto-oncogenes that participate in important 2
physiological functions such as cellular differentiation, mobility, and proliferation.
3
Src, a member of the Src-family protein kinases, functions as a signal protein and is 4
implicated in various diseases. Src is ubiquitously expressed within the body, but 5
distribution levels vary depending on individual tissue and organs (1,2). The role of 6
Src in cancer was initially established by its overexpression in colon (3-7) and 7
breast cancer (8-20) and is now well established. Figure 1 illustrates the major 8
cancer types in which Src plays a significant role, and important substrates and 9
factors involved in each pathway (21-39). The critical role of Src in cancer makes it 10
an attractive target for designing novel cancer therapy.
11
The carboxyl terminal of the Src kinase (CSK) is important in regulating 12
conformation and activity of Src. Under normal inactive conditions, the Src protein 13
is locked as an inward-folding conformation through binding between the 14
phosphorylated Tyr527 and the SH2 domain. However, in mitotic cells, Src 15
becomes activated due to thedephosphorylation of Tyr527 and phosphorylation of 16
Tyr419.
17
From N-terminal to C-terminal, Src is composed of a smaller amino-terminal 18
lobe (residues 270-340) which binds ATP, and a larger carboxyl-terminal lobe 19
(residues 345-523) which binds with substrates. The ATP binding site is also 20
partially located in the larger lobe. By regulating the structure of the alpha-helix, the 21
large lobe can move toward or away from the small lobe, resulting in opening or 22
closing of the cleft between the two lobes. The Src catalytic site is located within 23
the cleft. An open conformation allows the entrance of ATP into the cleft and exit of 24
ADP from the cleft. Drugs that can either interact with the residues (404-432) on the 25
4
activation loop or inhibit the activation loop from moving away and opening the 26
cleft as a result of Tyr419 phosphorylation can effectively inhibit Src activity.
27
The importance of Src is established in cancer, but increasing studies are also 28
suggesting its possible involvement in hypertension (40).Based on the 29
pathophysiology of hypertension, endogenous ouabain (EO) and mutant α-adducin 30
have been shown to increase binding of Na-K-ATPase with the SH2 domain and 31
phosphorylation of Tyr416, thus activating subsequent 32
Na,K-ATPase-Src-EGFR-ERK1/2 signaling cascade (41-43). As Src kinase is a key 33
protein regulating the activation of downstream pathways, inhibiting Src kinase 34
activation and Tyr416 phosphorylation may be a novel approach to controlling 35
hypertension. Observations showing that lung cancer patients also have the 36
tendency to develop hypertension and other cardiovascular diseases (44) also 37
suggest possible linkage between the two diseases. In view of this, we speculate that 38
Src may be a common factor for the development of hypertension and cancer, and 39
inhibiting Src may have multi-targeting effects for both diseases. Though many Src 40
inhibitors have been developed, limited pharmaceutical effectiveness has been 41
observed. Recently, computational techniques are commonly used to screen or 42
predict potent drugs targeting specific diseases (45-58).In view of this shift in drug 43
design, and considering the vast pharmaceutical potential of traditional Chinese 44
medicine (TCM), we constructed the world’s most comprehensive TCM database 45
TCM Database@Taiwan (http://tcm.cmu.edu.tw/) (59,60) and its accompanying 46
cloud-computing webserver iScreen (61) and integrative computational design 47
portal “integrated SysteMs Biology Associated Research with TCM” (iSMART) 48
(62) to facilitate drug development from TCM. TCM Database@Taiwan has been 49
successfully used to identify lead compounds for a variety of important diseases 50
5
(63-82). In our previous study, we utilized TCM Database@Taiwan to identify 51
potential Src inhibitors (83). Based on the essential role of Src in cancer, we deemed 52
it important to evaluate the effect of TCM compounds with Src inhibitory potential 53
against other cancer target proteins. This study targets not only Src, but also 54
validates the docking potential of these TCM candidates against established cancer 55
targets EGFR (84,85), HER2 (86,87), and HSP90 (88-91). Since Src is also 56
involved in the pathological mechanisms of EGFR and HER2 (92,93), ourTCM 57
candidates may have potential as multi-targeting inhibitors for different cancers.
58
Materials and Methods
59
Docking and Candidate Screening 60
The protein structure used in this study was downloaded from Protein Data 61
Bank (PDB: 2H8H) (94). The binding site was defined as the space occupied by 62
Saracatinib within the 2H8H crystal structure, and TCM compounds from TCM 63
Database@Taiwan docked and screened. Using Discovery Studio 2.5, the ligands 64
were first passed through Lipinski’s Rule of Five, and then screened for contour of 65
TCM ligands with the Src kinase binding site using LigandFit. LigandFit is a 66
receptor-rigid docking algorithm that uses Monte Carlo simulation to match ligands 67
with designate binding sites on a given protein. Results from docking are ranked by 68
binding energy and ligand similarity to the Saracatinib within the 2H8H crystal 69
structure, and three candidates are selected for further analysis. Each candidate 70
ligand was minimized with smart minimizer setting in Minimization Algorithm 71
under the force field of CHARMm (95), and a maximum of five docking poses were 72
generated. Each generated docking pose was redocked into the protein for a second 73
time, and five poses from each re-dock were generated. The 25 poses for each 74
ligand are then visually compared with that of Saracatinib within the crystal 75
6
structure, and docking poses that are most similar to Saracatinib were selected.
76
Hydrophobic contacts between the ligands and Src kinase are calculated with 77
LigPlot v.2.2.25(96). The absorption, distribution, metabolism, excretion and 78
toxicity (ADMET) analysis in D.S. 2.5 was used to calculate pharmacology and 79
toxicity of the derivatives in human bodies. To assess the general applicability of the 80
candidates on other cancer target proteins, the top three candidates from screening 81
were further docked into EGFR, HER2, and HSP. Protein structures used for this 82
application spectrum verification were EGFR (PDB: 2ITY) (97) and HSP90 (PDB:
83
3K97) (98). The HER2 model used was adopted from our previous study (72,99) 84
and was built from structures 2ITY and 2J5E (97,100).
85
Bioactivity Prediction by Multiple Linear Regression (MLR) and Support 86
Vector Machine (SVM) Models 87
Linear MLR (101) and nonlinear SVM (102) were used to construct 88
quantitative structure-activity relationship (QSAR) models for predicting the 89
bioactivity of the TCM compounds. A total of 20 Src ligands (103) were randomly 90
separated into a training set of 15 compounds and an external validation test set of 91
five compounds.
92
Prior to constructing QSAR models, genetic function approximation (GFA) 93
(104) was applied to identify representative descriptor sets from the large pool of 94
descriptors generated from the training set. The MLR model was built based on the 95
representative descriptors using MATLAB in the form of equation [1]:
96
n n n
x
pIC = +
∑
1 0
50 α α [1]
97
where α0 is a constant value and αn is the coefficient value of descriptor Xn.
98
The generated MLR model was validated with cross-validation and independent 99
7
tests and verified by calculating the square correlation coefficients (R2) between 100
predicted and actual pIC50 of the training set.
101
SVM are groups of supervised methods that allow categorization of hard-to 102
separate patterns through the use nonlinear of generalized portrait algorithms (105).
103
The SVM algorithm was extended for regression (SVMR) from its original use for 104
classification through the use of a ε-insensitive loss function (106) with the goal of 105
identifying a function f(x) in which all training points has a maximum deviation ε 106
from experimental values and has a maximum margin (107). A final nonlinear SVM 107
regression giving the modeled property for a pattern x was obtained by introducing 108
Lagrange multipliers and kernels to map input patterns into a higher dimension 109
space, the formula being equation [2]:
110
( )
K(
x x)
bx f
m
i
k i i i
k =
∑
− +=
+
− 1
, )
( λ λ
[2]
111
whereλ−i,λ+i are Lagrange multipliers and K(xi, xk) is the kernel function.
112
Our SVM model was constructed using the LibSVM program(108). Key 113
parameters determining the SVM model fit are C cost, ε, γ, the kernel type, and the 114
corresponding kernel parameters. The kernel selected for training the SVM model 115
was the Gaussian radial basis function kernel equation [3]:
116
( )
−
−
= 2
2
exp 2
, σ
y x x
x
K i k [3]
117
Optimum C, ε, γ were determined using the gridregression.py command within 118
LibSVM. Cross-validation was conducted according to default settings in LibSVM.
119
The validated MLR and SVM models to the TCM candidates to predict individual 120
pIC50 values of the compounds.
121
3D-QSAR Modeling and Analysis 122
8
3D-QSAR methods such as comparative molecular field analysis (CoMFA) 123
(109) and comparative molecular similarity analysis (CoMSIA) (110) are widely 124
used as activity prediction tools in drug design. Since traditional QSAR (MLR and 125
SVM) do not take into account the 3D structure of the compounds, CoMFA and 126
CoMSIA models were constructed to further test the robustness of TCM candidates 127
as ligands with biological activity against Src. The Src ligands from (103) were 128
randomly divided into a training set of 15 compounds and an external validation test 129
set of five compounds.
130
Partial least square (PLS) analysis is a statistical tool used for establishing a 131
linear model describing the correlation between dependent and independent 132
variables and has the advantage of being directly applicable for prediction(111). In 133
this section, CoMFA and CoMSIA descriptors were used as independent variables 134
and pIC50 values were the dependent variables.
135
The cross-validated coefficient, q2, which is calculated by equation [4], was used to 136
evaluate the prediction accuracy:
137
( )
( )
∑
∑
−
− −
= 2
_ 50 _
50
2 _ 50 _
2 1 50
mean actual
actual predicted
pIC pIC
pIC
q pIC [4]
138
Conventional correlation coefficient r2 and the standard error, SEE, were also 139
computed for each PLS model. Models with the highest q2, r2, and lowest SEE were 140
selected as the optimum CoMFA and CoMSIA model. Once the optimum model is 141
established, relevant compound descriptors are projected into the PLS model to 142
make external predictions on the test set.
143
CoMFA and CoMSIA structure building was performed using the SYBYL 144
program. The CoMFA descriptors steric and electrostatic field energies were 145
9
calculated by Lennard Jones function (112) and Coulombic function (113), 146
respectively, using the SYBYL default parameters: avan der Waals (vdW) radius of 147
1.52 Å, a C1+ probe atom, grid point spacing of 2Å, and energy cut-off value of 30 148
kcal/mol. The field contributions in CoMSIA, namely steric, electrostatic, 149
hydrophobic, H-bond donor and H-bond acceptor descriptors, were calculated with 150
identicalC1+ probe atom and grid spacing parameters. The probe radius was set at 151
1.0 Å. The default attenuation factor (R) value of 0.3 was used. Column filtering 152
was set at2.0 kcal/mol. The TCM ligands were overlaid against the generated 153
models to evaluate biological activities based on the CoMFA and CoMSIA models.
154
Molecular Dynamics (MD) Simulation 155
To verify the stability of TCM candidates under dynamic conditions, molecular 156
dynamics (MD) simulations were conducted on the Src-candidate complexes using 157
DS2.5. The energy of each complex was minimized with 500 steps each of Steepest 158
Descent and Conjugate Gradient. The system was heated for 50 ps to increase 159
temperature from 50K to 310K and allowed to equilibrate for 200 ps once the target 160
temperature was reached. Canonical ensemble (NVT; constant temperature) was 161
selected for the 40 ns production process and snapshots were taken at 20 ps intervals.
162
Time steps were set at 2fs. Electrostatic interactions (114) were calculated using 163
Particle Mesh Ewald (PME) method. The MD results were used to analyze energy 164
trajectories, H-bond formation and distances, and torsion angles which provide 165
insights to the interaction between TCM candidates and Src kinase.
166
Results and Discussion
167
Docking and Candidate Screening 168
. Table I lists the top ten TCM candidates with the lowest Binding Energy and 169
10
Ligand Internal Energy. The complete list of top 100 TCM ligands based on Binding 170
Energyand Ligand Internal Energy can be viewed in Supplementary Table I.
171
Considering the ability to form multiple bonds with Src and their chemical structure, 172
Angeliferulate,(3R)-2'-hydroxy-3',4'-dimethoxyisoflavan-7-O-beta-D-glucoside 173
(HMID), and 3-[2',6-dihydroxy-5'-(2-propenyl)[1,1'-biphenyl]3-yl]-(E)-2-propenoic 174
acid (3PA) were selected as our top candidates. The structural scaffolds of the TCM 175
candidates and the control Saracatinib are shown in Figure 2.Significantly lower 176
binding energy and LIE values were estimated for the TCM candidates (Table I).A 177
higher binding energy, such as the case with Saracatinib, implies that the ligand 178
binding with the protein is more unstable. Results for ADMET are shown in 179
Supplementary Figure 1.
180
Figures 3-6 illustrates interactions and Src protein residues that may be of 181
importance for each ligand. Figure 7 highlights amino acids that form hydrophobic 182
contacts with each test ligand. Saractatinib formed two types of interactions with 183
Src. Pi-interaction was formed with the positively charged alkyl group in Lys298 184
(Figure 3A), and hydrophobic interactions were formed with amino acids located on 185
different loops (Figure 7A). The low calculated binding energy and LIE for 186
Angeliferulate (Top 1) could be the result of H-bonds formed with Lys298, Asp351, 187
Asp407 (Figure 4A), and hydrophobic interactions with Val284, Gly347, and 188
Leu396 (Figure 7B). HMID is docked within Src through H-bonds with Leu276, 189
Lys298, and Asp407 (Figure 5A) and hydrophobic interactions with seven amino 190
acids (Figure 7C). These interactions are formed with amino acids either on the 191
11
activation loop (Ala407, Asp407) or on different loops (Leu276, Ser348), helping to 192
maintain the stability of the cleft. Similarly, 3PA also docked in Src through the 193
formation of H-bonds (Lys298, Ser348, and Asp351) (Figure 6A) and hydrophobic 194
interactions formed with amino acids on different loops (Figure 7D).
195
Lys298 is an amino acid of critical importance in Src activation. Under open 196
conformations, ATP will enter the cleft, bind atLys298, undergo hydrolysis to 197
release a phosphate and exit as ADP. The released phosphate is used to 198
phosphorylate downstream reactions. A ligand that can effectively interact with 199
Lys298and form stable and permanent interactions with neighboring amino acid 200
residues has potential to block ATP from the binding site and inhibit activation of 201
downstream reactions. Whilst Saracatinib formed a pi-interaction with Lys298, lack 202
of H-bonds with other residues renders Saracatinib unstable. Higher stability of 203
TCM candidates over Saracatinib may be likely due to their ability to form H-bonds 204
with multiple amino acids including Lys298. Hydrophobic contacts, through 205
significantly weaker than pi-interactions and H-bonds, may also contribute to 206
stability. In particular, hydrophobic interactions formed by Saracatinib, HMGF, and 207
3PA with Leu276 and Ser348 (located on different amino acid chains) (Figure 7) 208
can increase stability of the cleft. HMID also forms hydrophobic interactions with 209
activation loop amino acids Ala406 and Asp407 which can resist removal of the 210
activation loop from the cleft during Tyr419 activation.
211
Bioactivity Prediction by Multiple Linear Regression (MLR)and Support 212
Vector Machine (SVM) Models 213
The MLR and SVM models were developed with ten structural descriptors 214
identified by GFA, namely ALogP, ES_Sum_dsCH, ES_Sum_aaCH, ES_Sum_sCl, 215
Molecular_Weight, Molecular_SurfaceArea, Molecular_PolarSurfaceArea, 216
12
Jurs_RPCG, Jurs_WNSA_1, and PMI_X.ALogPis a measurement of molecular 217
hydrophobicity based on Ghose and Crippen’s method (115); ES_Sum_dsCHis 218
theelectrotopological count for carbons with one single bond and one double 219
bond,ES_Sum_aaCH stand for the electrotopological count for carbons with two 220
aromatic bonds; ES_Sum_sCl is related to the electrotopological count of chlorides 221
with a single bond; Molecular_Weight is the sum of atomic weight;
222
Molecular_SurfaceArea and Molecular_PolarSurfaceAreacalculate total surface 223
area and polar surface area, respectively; Jurs_RPCGdescribes relative positive 224
charge; Jurs_WNSA_1describes the total molecular solvent-accessible surface, and 225
PMI_X is a spatial descriptor related to the orientation and conformational rigidity 226
of the ligand.
227
The generated MLR model is expressed as follows and has good prediction ability 228
(R2= 0.8043; Figure 8A):
229
X PMI WNSA
Jurs
RPCG Jurs
ceArea PolarSurfa
Molecular
a SurfaceAre Molecular
Weight Molecular
sCl Sum ES aaCH
Sum ES
dsCH Sum
ES ALogP
pIC
_ 0060
. 0 1 _ _
0050 . 0
_ 3521
. 7 _
0567 . 0
_ 0107
. 0 _
0433 . 0
_ _ 9470 . 33 _
_ 879 . 0
_ _ 2497 . 9 2414
. 0 7 .
50 387
× +
× +
×
−
×
−
× +
×
+
×
−
×
−
× +
×
−
=
230
The SVM model constructed using the aforementioned descriptors also 231
generated a prediction model where predicted values were highly correlated to 232
actual observed values (R2=0.937; Figure 8B). The robustness of our models were 233
validated through external validation tests using the test set. Good correlation 234
between observed and predicted pIC50 values were observed for both models.
235
Predicted pIC50values of the TCM candidates and Saracatinib using the MLR 236
and SVM models are listed in Table I. Results suggest that the TCM candidates 237
have good bioactivity towards Src.
238
3D-QSAR Modeling and Analysis 239
13
Table II shows the results of the CoMFA and CoMSIA models generated 240
through PLS algorithm. PLS statistics led to a CoMFA model in which steric 241
features were the dominant contributing factor. Several CoMSIA model 242
combinations were generated and the model considering electrostatic (E), 243
hydrophobic (H), and H-bond donor (D) properties was selected as the optimum 244
model based on high cross-validation and non-cross validation correlation 245
coefficients (q2=0.482, r2=0.877).
246
To test the predictive capabilities of the models, they were used to predict the 247
pIC50 values of an external test set excluded from the original training set.
248
Bioactivity predictions of the training and test set are listed in Table III.The 249
correlation curves show high square correlation coefficients of R2=0.9721 for 250
CoMFA (Figure 9A) and R2=0.9414 for CoMSIA (Figure 9B), implying models of 251
good prediction power.
252
In Figure 10, the heteroaromatic ring of Saracatinib, the benzene of 253
Angeliferulate, and the hydroxyl group of HMID, and partial regions of the hexane 254
ring in 3PA fall within the steric favor region, but only Angeliferulate forms 255
interaction bonds. All test ligands fall within the regions between Leu276/Asp351 256
and Lys298/Asp407, satisfying the CoMFA steric favoring region located at 257
Leu276/Asp351. However, only TCM candidates have interactions with the amino 258
acids located within the Lys298/Asp407 region. When the CoMSIA contour was 259
superimposed, Lys298 was located within the region where hydrophilic interactions 260
were desirable for bioactivity (white)(Figure 11). The ability of Angeliferulate, 261
HMID and 3PA to form H-bonds at Lys298 contour to this hydrophilic region and 262
most likely contribute to higher bioactivities. The hydroxyl group of Angeliferulate 263
falls adjacent to the electrostatic favoring (orange) region of Leu276/Asp351 264
14
(Figure 11B). The hydrophobic benzene ring moiety of 3PA is located close to the 265
hydrophobic disfavoring region (white) of Leu276/Asp351 (Figure 11D). Violation 266
of the hydrophobic disfavor region contour matches the lower bioactivity predicted 267
by our SVM/MLR models (Table I).
268
Molecular Dynamics (MD) Simulation 269
RMSDs and total energy trajectories 270
Protein-ligand complex RMSDs, ligand RMSDs, and total energy level 271
changes during MD are shown in Figure 12. Saracatinib has the highest complex, 272
ligand RMSDs, and the highest total energy of the four test compounds. Most 273
notably, the significant increase in ligand RMSD (Figure 12B) indicates high 274
instability. Total energy of the TCM candidates by increasing order 275
were3PA<Angeliferulate<HMID, all of which were lower than that of Saracatinib 276
(Figure 12C). The lower total energy trajectories indicate a more stable state of the 277
ligand-protein complex during dynamic situations.
278
Saracatinib-Src interactions during MD 279
Saracatinib formed the least amount of interactions with Src during docking, a 280
phenomenon also observed during MD (Figure 3A).As indicated by the H-bond 281
distance trajectories (Figure 3B) and H-bond occupancy analysis (Table IV), the 282
only stable H-bonds formed during MD were with Lys298 and Ser348. Saracatinib 283
could not form H-bonds with neighboring residues Leu276, Asp351, and Asp407 284
were the underlying reasons for the inability to form H-bonds.
285
Angeliferulate-Src interactions during MD 286
The primary stabilizing interactions formed between Angeliferulate and Src 287
were at Ser348, Asp351, and Asp407 (Figure 4B). H-bonds with Lys298 were also 288
initially observed, but rotations on Angeliferulate increased the distance and 289
15
discouraged interactions with Lys298 after 9.86 ns. Distance from Asp351 reduced 290
from 4.47Å to 2.45Å at 7.64ns, enabling the formation of multiple H-bonds. Small 291
fluctuations on the H-bond distanceswithSer348 and Asp407 were observed, but 292
since the distance was greater than the 2.50Å cut-off distance, the H-bonds were 293
presumed to be weaker.
294
HMID-Src interactions during MD 295
A significant directional shift of HMID was observed during MD due to the 296
formation of H-bonds with new amino acid residues (Table IV). Initially, the 297
H-bond with Leu276 during directed HMID towards the small lobe (Figure 5A).
298
During MD, the loss of the Leu276 H-bond and the formation of H-bond with 299
Ser348 (Figure 5B) shifted HMID towards the large lobe (Figure 5A). HMID 300
formed seven H-bonds with Lys298, but as the HZ1, HZ2, and HZ3 on Lys298 301
continuously rotated, only a 50.65% occupancy was recorded. The H-bond with 302
Asp407 observed during docking was stable throughout MD.
303
3PA-Src interactions during MD 304
3PA was primarily stabilized by Lys298 and Asp351 during MD (Table IV).
305
Multiple H-bonds were formed with Lys298. Distance fluctuations recorded in 306
Figure 6B were normal circumstances caused by the constant rotation of the Lys298 307
H atoms. A stable H-bond with occupancies of greater than 99% was maintained 308
with Asp351. Contrary to the docking pose, no high occupancy H-bonds were 309
observed for Ser348.
310
Torsion angle changes during MD 311
More information on bond formation during MD can be explained through 312
torsion angle changes. As shown in Figure 3C, torsion of a and b contribute to the 313
relative spatial angle of the chloride-containing moiety of Saracatinib. The inability 314
16
of Saracatinib to maintain interactions with Lys298 and Asp407 could be attributed 315
to conformational shifts brought on by these torsions which increase the distance of 316
these residues from Saracatinib. Oxane could not form interactions due to large 317
torsion fluctuations in c and d and its orientation away from the pocket. Piperazine 318
groups could not form H-bonds due to the large torsion changes observed in e, f, g, 319
and h.
320
Angeliferulate was a largely flexible ligand as indicated by the torsion angles 321
(Figure 4C). Large torsion changes at a-d disrupted the ability of the terminal 322
aromatic ring to form stable H-bonds. Torsion angles at e-h also showed that 323
constant fluctuations. Though m and n were more stable, the dynamic outer regions 324
of the cleft limited the methoxy group and hydroxyl groups of m and n from 325
forming H-bonds.
326
Fluctuations of the torsion angles were also observed in HMID, but H-bond 327
formations were rarely affected (Figure 5C). The stability of b, e, and f enabled the 328
formation of stable H-bonds with Lys298 and Asp407 through O17 and O28, 329
respectively. Other recorded fluctuations at a-d, and g did not affect stability of 330
these two H-bonds. Two large torsion changes were observed at h (8.98 ns and 15 331
ns), both of which directly caused H-bond distance changes between H55 and 332
Leu276, Ser348, and Asp351 (Figure 5B).
333
Figure 6C shows that c, which connects the two benzene moieties in 3PA were 334
very stable. Torsions at e, f, g did not fluctuate greatly, resulting in the stable 335
H-bond with Lys298. 3PA remained in a relatively linear conformation until changes 336
observed at d. Nonetheless, aromatic rings remained in a planar state after 22 ns.
337
Discussion 338
Based on the results of docking and MD, Leu276, Lys298, Ser348, Asp351, 339
17
and Asp407 are important amino acid residues for stability within Src (Table V).
340
MD results generally supported the findings of docking but showed additional 341
H-bond formation. This implies that under dynamic physiological conditions, all 342
ligands tested could form and remain in complex with Src. Comparison between 343
ligands show that the TCM candidates were more stable and had lower total energy 344
than Saracatinib. The higher stability of the TCM candidates can be attributed to the 345
formation of more H-bonds that remain stable throughout MD. Angeliferulate and 346
HMID not only formed H-bond with the ATP binding site Lys298, but also with 347
Asp407. This can effectively limit the phosphorylated Tyr419 from moving away 348
from the cleft, thus exhibiting important characteristics for being potential Src 349
inhibitors.
350
Conclusion
351
This study utilizes computational methods to virtually screen small molecules 352
found in TCM for potential Src inhibitors. Potential of each candidate was validated 353
using structure-based and ligand-based drug design. As shown in Supplementary 354
Video 1, Angeliferulate and HMID have multiple stable interactions with the two 355
Src cleft loops while simultaneously interacting with Asp407, hindering the 356
activation loop from activation. 3PA also exhibits drug-like potential by primarily 357
interacting with Src through via cleft loop amino acids, but may be less potent than 358
its TCM counterparts due to the lack of direct interaction with the activation loop.
359
Considering the aforementioned interactions with Src and high affinity with EGFR, 360
HER2, and HSP90, we suggest that Angeliferulate and HMID which both originate 361
from the TCM Angelica sinensis may have potential as multi-targeting drug leads.
362
Supplementary Materials
363
Supplementary material include the top 100 candidates from screening, ADMET 364
18
results of the top three TCM candidates, and a video depicting the mode of 365
inhibition of the TCM candidates on Src.
366
Acknowledgements
367
The research was supported by grants from the National Science Council of 368
Taiwan (NSC 100-2325-B-039-001), Committee on Chinese Medicine and 369
Pharmacy (CCMP100-RD-030), China Medical University and Asia University 370
(DMR-101-094). This study is also supported in part by Taiwan Department of 371
Health Clinical Trial and Research Center of Excellence (DOH101-TD-B-111-004) 372
and Taiwan Department of Health Cancer Research Center of Excellence 373
(DOH101-TD-C-111-005). We are grateful to the National Center of 374
High-performance Computing for computer time and facilities and Dr. Su-sen 375
Chang for discussions and technical assistance with the manuscript preparations. We 376
also wish to express thanks to cloud-computing facilities at Asia University.
377
19
References
378
1. G. Manning, D. B. Whyte, R. Martinez, T. Hunter, and S. Sudarsanam.
379
Science 298, 1912-1934 (2002).
380
2. S. M. Thomas and J. S. Brugge. Annu Rev Cell Dev Biol 13, 513-609 (1997).
381
3. C. Oneyama, E. Morii, D. Okuzaki, Y. Takahashi, J. Ikeda, N. Wakabayashi, 382
H. Akamatsu, M. Tsujimoto, T. Nishida, K. Aozasa, and M. Okada.
383
Oncogene, (Advanced Online Publication) doi: 10.1038/onc.2011.1367 384
(2011).
385
4. J. Huang, L. Yao, R. Xu, H. Wu, M. Wang, B. S. White, D. Shalloway, and 386
X. Zheng. Embo J 30, 3200-3211 (2011).
387
5. G. Swaminathan and C. A. Cartwright. Oncogene, (Advanced Online 388
Publication) doi:10.1038/onc.2011.1242 (2011).
389
6. R. R. Brady, C. J. Loveridge, M. G. Dunlop, and L. A. Stark.
390
Carcinogenesis 32, 1069-1077 (2011).
391
7. D. Gianni, N. Taulet, C. DerMardirossian, and G. M. Bokoch. Mol Biol Cell 392
21, 4287-4298 (2010).
393
8. B. Pohorelic, R. Singh, S. Parkin, K. Koro, A. D. Yang, C. Egan, and A.
394
Magliocco. Breast Cancer Res Tr, (Advanced Online Publication) doi:
395
10.1007/s10549-10011-11753-10542 (2011).
396
9. J. Jian, Q. Yang, and X. Huang. J Biol Chem 286, 35708-35715 (2011).
397
10. A. J. Bernier, J. Zhang, E. Lillehoj, A. R. Shaw, N. Gunasekara, and J. C.
398
Hugh. Mol Cancer 10, 93 (2011).
399
11. V. Ratushny, H. B. Pathak, N. Beeharry, N. Tikhmyanova, F. Xiao, T. Li, S.
400
Litwin, D. C. Connolly, T. J. Yen, L. M. Weiner, A. K. Godwin, and E. A.
401
Golemis. Oncogene, doi: 10.1038/onc.2011.1314 (2011).
402
12. N. Tikhmyanova and E. A. Golemis. PLoS One 6, e22102 (2011).
403
13. T. Miyake and S. J. Parsons. Oncogene, doi: 10.1038/onc.2011.1332 (2011).
404
14. M. E. Irwin, N. Bohin, and J. L. Boerner. Cancer Biol Ther 12, 718-726 405
(2011).
406
15. T. J. Shackleford, Q. Zhang, L. Tian, T. T. Vu, A. L. Korapati, A. M.
407
Baumgartner, X. F. Le, W. S. Liao, and F. X. Claret. Breast Cancer Res 13, 408
R65 (2011).
409
16. J. D. Bjorge, A. S. Pang, M. Funnell, K. Y. Chen, R. Diaz, A. M. Magliocco, 410
and D. J. Fujita. PLoS One 6, e19309 (2011).
411
17. Z. Hou, D. J. Falcone, K. Subbaramaiah, and A. J. Dannenberg.
412
Carcinogenesis 32, 695-702 (2011).
413
18. C. C. Mader, M. Oser, M. A. Magalhaes, J. J. Bravo-Cordero, J. Condeelis, 414
A. J. Koleske, and H. Gil-Henn. Cancer Res 71, 1730-1741 (2011).
415
20
19. K. K. Haenssen, S. A. Caldwell, K. S. Shahriari, S. R. Jackson, K. A.
416
Whelan, A. J. Klein-Szanto, and M. J. Reginato. J Cell Sci 123, 1373-1382 417
(2010).
418
20. K. E. Reeder-Hayes, L. A. Carey, and W. M. Sikov. Breast Dis 32, 123-136 419
(2010).
420
21. S. Thomas, J. B. Overdevest, M. D. Nitz, P. D. Williams, C. R. Owens, M.
421
Sanchez-Carbayo, H. F. Frierson, M. A. Schwartz, and D. Theodorescu.
422
Cancer Res 71, 832-841 (2011).
423
22. N. Said and D. Theodorescu. Cancer Metastasis Rev 28, 327-333 (2009).
424
23. A. Prinetti, T. Cao, G. Illuzzi, S. Prioni, M. Aureli, N. Gagliano, G. Tredici, 425
V. Rodriguez-Menendez, V. Chigorno, and S. Sonnino. J Biol Chem 47, 426
40900-40910 (2011).
427
24. S. Charoenfuprasert, Y. Y. Yang, Y. C. Lee, K. C. Chao, P. Y. Chu, C. R.
428
Lai, K. F. Hsu, K. C. Chang, Y. C. Chen, L. T. Chen, J. Y. Chang, S. J. Leu, 429
and N. Y. Shih. Oncogene 30, 3570-3584 (2011).
430
25. H. S. Kim, H. D. Han, G. N. Armaiz-Pena, R. L. Stone, E. J. Nam, J. W. Lee, 431
M. M. Shahzad, A. M. Nick, S. J. Lee, J. W. Roh, M. Nishimura, L. S.
432
Mangala, J. Bottsford-Miller, G. E. Gallick, G. Lopez-Berestein, and A. K.
433
Sood. Clin Cancer Res 17, 1713-1721 (2011).
434
26. E. L. Leung, J. C. Wong, M. G. Johlfs, B. K. Tsang, and R. R. Fiscus. Mol 435
Cancer Res 8, 578-591 (2010).
436
27. X. G. Liu, Y. Guo, Z. Q. Yan, M. Y. Guo, Z. G. Zhang, and C. A. Guo.
437
Zhonghua Zhong Liu Za Zhi 33, 340-344 (2011).
438
28. Y. Ding, X. Wang, A. Xu, X. Xu, K. Tian, C. Y. Young, and H. Yuan. J Cell 439
Biochem 112, 818-828 (2011).
440
29. T. Kobayashi, T. Inoue, Y. Shimizu, N. Terada, A. Maeno, Y. Kajita, T.
441
Yamasaki, T. Kamba, Y. Toda, Y. Mikami, T. Yamada, T. Kamoto, O.
442
Ogawa, and E. Nakamura. Mol Endocrinol 24, 722-734 (2010).
443
30. S. Zhang, H. E. Zhau, A. O. Osunkoya, S. Iqbal, X. Yang, S. Fan, Z. Chen, 444
R. Wang, F. F. Marshall, L. W. Chung, and D. Wu. Mol Cancer 9, 9 (2010).
445
31. J. DaSilva, D. Gioeli, M. J. Weber, and S. J. Parsons. Cancer Res 69, 446
7402-7411 (2009).
447
32. G. Pandini, M. Genua, F. Frasca, R. Vigneri, and A. Belfiore. Ann N Y Acad 448
Sci 1155, 263-267 (2009).
449
33. F. Leve, T. G. Marcondes, L. G. Bastos, S. V. Rabello, M. N. Tanaka, and J.
450
A. Morgado-Diaz. Eur J Pharmacol 671, 7-17 (2011).
451
34. L. M. Sturla, P. O. Zinn, K. Ng, M. Nitta, D. Kozono, C. C. Chen, and E. M.
452
Kasper. Brit J Cancer 105, 1235-1243 (2011).
453
21
35. A. E. Al Moustafa, A. Yasmeen, and A. Achkhar. Med Hypotheses 77, 454
812-814 (2011).
455
36. S. Saini, S. Arora, S. Majid, V. Shahryari, Y. Chen, G. Deng, S. Yamamura, 456
K. Ueno, and R. Dahiya. Cancer Prev Res 4, 1698-1709 (2011).
457
37. X. F. Le, W. Mao, G. He, F. X. Claret, W. Xia, A. A. Ahmed, M. C. Hung, 458
Z. H. Siddik, and R. C. Bast, Jr. J Natl Cancer I 103, 1403-1422 (2011).
459
38. A. Migliaccio, G. Castoria, and F. Auricchio. Methods Mol Biol 776, 460
361-370 (2011).
461
39. J. C. Lee, M. C. Maa, H. S. Yu, J. H. Wang, C. K. Yen, S. T. Wang, Y. J.
462
Chen, Y. Liu, Y. T. Jin, and T. H. Leu. Mol Carcinog 43, 207-214 (2005).
463
40. M. Ferrandi, I. Molinari, L. Torielli, G. Padoani, S. Salardi, M. P. Rastaldi, P.
464
Ferrari, and G. Bianchi. Sci Transl Med 2, 59ra86 (2010).
465
41. P. M. Kearney, M. Whelton, K. Reynolds, P. Muntner, P. K. Whelton, and J.
466
He. Lancet 365, 217-223 (2005).
467
42. P. Ferrari. Biochim Biophys Acta 1802, 1254-1258 (2010).
468
43. P. Ferrari, M. Ferrandi, G. Valentini, and G. Bianchi. Am J Physiol-Reg I 469
290, R529-R535 (2006).
470
44. P. C. Chen, C. H. Muo, Y. T. Lee, Y. H. Yu, and F. C. Sung. Stroke 42, 471
3034-3039 (2011).
472
45. A. P. Guimaraes, A. A. Oliveira, E. F. da Cunha, T. C. Ramalho, and T. C.
473
Franca. J Biomol Struct Dyn 28, 455-469 (2011).
474
46. E. P. Semighini, J. A. Resende, P. de Andrade, P. A. B. Morais, I. Carvalho, 475
C. A. Taft, and C. H. T. P. Silva. J Biomol Struct Dyn 28, 787-796 (2011).
476
47. Y. D. Cai, J. F. He, and L. Lu. J Biomol Struct Dyn 28, 797-804 (2011).
477
48. Z. H. Mei, J. Liu, and H. W. Yu. J Biomol Struct Dyn 28, 871-879 (2011).
478
49. K. C. Chen and C. Y. C. Chen. Soft Matter 7, 4001-4008 (2011).
479
50. H. J. Huang, K. J. Lee, H. W. Yu, H. Y. Chen, F. J. Tsai, and C. Y. C. Chen.
480
J Biomol Struct Dyn 28, 187-200 (2010).
481
51. C. Y. C. Chen. J Biomol Struct Dyn 27, 627-640 (2010).
482
52. K. Bhargavi, P. Kalyan Chaitanya, D. Ramasree, M. Vasavi, D. K. Murthy, 483
and V. Uma. J Biomol Struct Dyn 28, 379-391 (2010).
484
53. M. T. Cambria, D. Di Marino, M. Falconi, S. Garavaglia, and A. Cambria. J 485
Biomol Struct Dyn 27, 501-509 (2010).
486
54. G. Ompraba, D. Velmurugan, P. A. Louis, and Z. A. Rafi. J Biomol Struct 487
Dyn 27, 489-499 (2010).
488
55. E. F. F. da Cunha, E. F. Barbosa, A. A. Oliveira, and T. C. Ramalho. J 489
Biomol Struct Dyn 27, 619-625 (2010).
490
56. C. Y. C. Chen. J Taiwan Inst Chem E 41, 143-149 (2010).
491
22
57. C. Y. C. Chen. J Taiwan Inst Chem E 40, 55-69 (2009).
492
58. C. Y. C. Chen. J Chin Chem Soc-Taip 54, 653-658 (2007).
493
59. C. Y. C. Chen. PLoS One 6, e15939 (2011).
494
60. K. Sanderson. Nat Med 17, 1531 (2011).
495
61. T. Y. Tsai, K. W. Chang, and C. Y. C. Chen. J Comput Aid Mol Des 25, 496
525-531 (2011).
497
62. K. W. Chang, T. Y. Tsai, K. C. Chen, S. C. Yang, H. J. Huang, T. T. Chang, 498
M. F. Sun, H. Y. Chen, F. J. Tsai, and C. Y. C. Chen. J Biomol Struct Dyn 499
29, 243-250 (2011).
500
63. S. S. Chang, H. J. Huang, and C. Y. C. Chen. PLoS Comput Biol 7, 501
e1002315 (2011).
502
64. S. S. Chang, H. J. Huang, and C. Y. C. Chen. Mol Biosyst 7, 3366-3374 503
(2011).
504
65. K. C. Chen, M. F. Sun, S. C. Yang, S. S. Chang, H. Y. Chen, F. J. Tsai, and 505
C. Y. C. Chen. Chem Biol Drug Des 78, 679-688 (2011).
506
66. T. T. Chang, K. C. Chen, K. W. Chang, H. Y. Chen, F. J. Tsai, M. F. Sun, 507
and C. Y. C. Chen. Mol Biosyst 7, 2702-2710 (2011).
508
67. M. F. Sun, H. Y. Chen, F. J. Tsai, S. H. Lui, and C. Y. C. Chen. J Biomol 509
Struct Dyn 29, 325-337 (2011).
510
68. M. F. Sun, T. T. Chang, K. W. Chang, H. J. Huang, H. Y. Chen, F. J. Tsai, J.
511
G. Lin, and C. Y. C. Chen. J Biomol Struct Dyn 28, 895-906 (2011).
512
69. T. T. Chang, M. F. Sun, H. Y. Chen, F. J. Tsai, M. Fisher, J. G. Lin, and C.
513
Y. C. Chen. J Biomol Struct Dyn 28, 773-786 (2011).
514
70. C. H. Lin, T. T. Chang, M. F. Sun, H. Y. Chen, F. J. Tsai, K. L. Chang, M.
515
Fisher, and C. Y. C. Chen. J Biomol Struct Dyn 28, 471-482 (2011).
516
71. P. C. Chang, J. D. Wang, M. M. Lee, S. S. Chang, T. Y. Tsai, K. W. Chang, 517
F. J. Tsai, and C. Y. C. Chen. J Biomol Struct Dyn 29, 471-483 (2011).
518
72. S. C. Yang, S. S. Chang, and C. Y. C. Chen. PLoS One 6, e28793 (2011).
519
73. S. C. Yang, S. S. Chang, H. Y. Chen, and C. Y. C. Chen. PLoS Comput Biol 520
7, e1002189 (2011).
521
74. T. T. Chang, H. J. Huang, K. J. Lee, H. W. Yu, H. Y. Chen, F. J. Tsai, M. F.
522
Sun, and C. Y. C. Chen. J Biomol Struct Dyn 28, 309-321 (2010).
523
75. C. Y. Chen and C. Y. C. Chen. J Mol Graph Model 29, 21-31 (2010).
524
76. H. J. Huang, K. J. Lee, H. W. Yu, C. H. Hsu, H. Y. Chen , F. J. Tsai, and C.
525
Y. C. Chen. J Biomol Struct Dyn 28, 23-37 (2010).
526
77. C. Y. C. Chen. J Taiwan Inst Chem E 40, 155-161 (2009).
527
78. C. Y. Chen, Y. H. Chang, D. T. Bau, H. J. Huang, F. J. Tsai, C. H. Tsai, and 528
C. Y. C. Chen. Acta Pharmacol Sin 30, 1186-1194 (2009).
529
23
79. C. Y. C. Chen. J Chin Inst Chem Eng 39, 291-299 (2008).
530
80. C. Y. C. Chen. J Chin Inst Chem Eng 39, 663-671 (2008).
531
81. C. Y. C. Chen, G. W. Chen, and W. Y. C. Chen. J Chin Chem Soc-Taip 55, 532
297-302 (2008).
533
82. C. Y. Chen, H. J. Huang, F. J. Tsai, and C. Y. C. Chen. J Taiwan Inst Chem 534
E 41, 8-15 (2010).
535
83. W. I. Tou and C. Y. C. Chen. PLoS One In press. (2012).
536
84. H. W. Lo, S. C. Hsu, W. Xia, X. Cao, J. Y. Shih, Y. Wei, J. L. Abbruzzese, 537
G. N. Hortobagyi, and M. C. Hung. Cancer Res 67, 9066-9076 (2007).
538
85. H. W. Lo, S. C. Hsu, M. Ali-Seyed, M. Gunduz, W. Xia, Y. Wei, G.
539
Bartholomeusz, J. Y. Shih, and M. C. Hung. Cancer Cell 7, 575-589 (2005).
540
86. M. Casimiro, O. Rodriguez, L. Pootrakul, M. Aventian, N. Lushina, C.
541
Cromelin, G. Ferzli, K. Johnson, S. Fricke, F. Diba, B. Kallakury, C.
542
Ohanyerenwa, M. Chen, M. Ostrowski, M. C. Hung, S. A. Rabbani, R. Datar, 543
R. Cote, R. Pestell, and C. Albanese. Cancer Res 67, 4364-4372 (2007).
544
87. C. Bartholomeusz, H. Itamochi, L. X. Yuan, F. J. Esteva, C. G. Wood, N.
545
Terakawa, M. C. Hung, and N. T. Ueno. Cancer Res 65, 8406-8413 (2005).
546
88. Y. Yufu, J. Nishimura, and H. Nawata. Leukemia Res 16, 597-605 (1992).
547
89. M. Ferrarini, S. Heltai, M. R. Zocchi, and C. Rugarli. Int J Cancer 51, 548
613-619 (1992).
549
90. A. Jameel, M. Law, R. Coombes, and Y. Luqmani. Int J Oncol 2, 1075-1080 550
(1993).
551
91. L. Whitesell, E. G. Mimnaugh, B. De Costa, C. E. Myers, and L. M. Neckers.
552
P Natl Acad Sci U S A 91, 8324-8328 (1994).
553
92. T. H. Leu and M. C. Maa. Front Biosci 8, s28-38 (2003).
554
93. E. M. Poole, K. Curtin, L. Hsu, R. J. Kulmacz, D. J. Duggan, K. W. Makar, 555
L. Xiao, C. S. Carlson, M. L. Slattery, B. J. Caan, J. D. Potter, and C. M.
556
Ulrich. Int J Mol Epidemiol Genet 2, 300-315 (2011).
557
94. L. F. Hennequin, J. Allen, J. Breed, J. Curwen, M. Fennell, T. P. Green, C.
558
Lambert-van der Brempt, R. Morgentin, R. A. Norman, A. Olivier, L.
559
Otterbein, P. A. Ple, N. Warin, and G. Costello. J Med Chem 49, 6465-6488 560
(2006).
561
95. B. R. Brooks, C. L. Brooks, 3rd, A. D. Mackerell, Jr., L. Nilsson, R. J.
562
Petrella, B. Roux, Y. Won, G. Archontis, C. Bartels, S. Boresch, A. Caflisch, 563
L. Caves, Q. Cui, A. R. Dinner, M. Feig, S. Fischer, J. Gao, M. Hodoscek, 564
W. Im, K. Kuczera, T. Lazaridis, J. Ma, V. Ovchinnikov, E. Paci, R. W.
565
Pastor, C. B. Post, J. Z. Pu, M. Schaefer, B. Tidor, R. M. Venable, H. L.
566
Woodcock, X. Wu, W. Yang, D. M. York, and M. Karplus. J Comput Chem 567
24
30, 1545-1614 (2009).
568
96. R. A. Laskowski and M. B. Swindells. J Chem Inf Model 51, 2778-2786 569
(2011).
570
97. C. H. Yun, T. J. Boggon, Y. Li, M. S. Woo, H. Greulich, M. Meyerson, and 571
M. J. Eck. Cancer Cell 11, 217-227 (2007).
572
98. P. P. Kung, B. Huang, G. Zhang, J. Z. Zhou, J. Wang, J. A. Digits, J.
573
Skaptason, S. Yamazaki, D. Neul, M. Zientek, J. Elleraas, P. Mehta, M. J.
574
Yin, M. J. Hickey, K. S. Gajiwala, C. Rodgers, J. F. Davies, and M. R.
575
Gehring. J Med Chem 53, 499-503 (2010).
576
99. M. F. Sun, S. C. Yang, K. W. Chang, T. Y. Tsai, H. Y. Chen, F. J. Tsai, J. G.
577
Lin, and C. Y. C. Chen. Mol Simulat 37, 884-892 (2011).
578
100. J. A. Blair, D. Rauh, C. Kung, C. H. Yun, Q. W. Fan, H. Rode, C. Zhang, M.
579
J. Eck, W. A. Weiss, and K. M. Shokat. Nat Chem Biol 3, 229-238 (2007).
580
101. B. K. Slinker and S. A. Glantz. Circulation 117, 1732-1737 (2008).
581
102. R. Burbidge, M. Trotter, B. Buxton, and S. Holden. Comput Chem 26, 5-14 582
(2001).
583
103. D. H. Boschelli, D. Wang, Y. Wang, B. Wu, E. E. Honores, A. C. Barrios 584
Sosa, I. Chaudhary, J. Golas, J. Lucas, and F. Boschelli. Bioorg Med Chem 585
Lett 20, 2924-2927 (2010).
586
104. M. Tufail and L. E. Ormsbee. J Hydroinform 8, 193-206 (2006).
587
105. V. Vapnik and A. Lerner. Automat Remote Contr 24, 774-780 (1963).
588
106. V. Vapnik. The Nature of Statistical Learning Theory. (Springer, 1995).
589
107. O. Ivanciuc. Reviews in Computational Chemistry 23, 291-400 (2007).
590
108. C.-C. Chang and C.-J. Lin. ACM TSIT, 2:27:21--27:27 (2011).
591
109. R. D. Cramer, 3rd, D. E. Patterson, and J. D. Bunce. Prog Clin Biol Res 291, 592
161-165 (1989).
593
110. G. Klebe, U. Abraham, and T. Mietzner. J Med Chem 37, 4130-4146 (1994).
594
111. T. Langer and S. D. Bryant. in The Practice of Medicinal Chemistry (ed 595
C. G. Wermuth) Ch. 29, 587-604 (Path Press, 2008).
596
112. R. J. Gillespie, D. Bayles, J. Platts, G. L. Heard, and R. F. W. Bader. J Phys 597
Chem A 102, 3407-3414 (1998).
598
113. G. Tang and B. K. Ma. Chinese Phys 9, 737-741 (2000).
599
114. M. Kawata and M. Mikami. Chem Phys Lett 313, 261-266 (1999).
600
115. A. K. Ghose and G. M. Crippen. J Chem Inf Comput Sci 27, 21-35 (1987).
601 602 603
