From virtual screening to bioactive compounds by visualizing and clustering of chemical space

Full text of DOI:10.1002/minf.201100147

DOI: 10.1002/minf.201100147
From Virtual Screening to Bioactive Compounds by
Visualizing and Clustering of Chemical Space
Alexander Klenner,[a] Volker Hꢀhnke,[a] Tim Geppert,[a] Petra Schneider,[a] Heiko Zettl,[a] Sarah Haller,[a]
Tiago Rodrigues,[a] Felix Reisen,[a] Benjamin Hoy,[b] Anja Maria Schaible,[c] Oliver Werz,[c] Silja Wessler,[b] and
Gisbert Schneider*[a]
Keywords: Drug discovery · Helicobacter pylori · Enzyme inhibitor · Pseudoreceptor · Target prediction · Virtual screening
Identification and visualization of ‘activity islands’ in chemi-
cal space is presented as a straightforward method for
rapid automated identification of bioactive compounds and
drug target profiling. We successfully applied this computa-
tional technique to finding inhibitors of Helicobacter pylori
protease HtrA with new molecular scaffolds, and to
deorphanizing of a compound from a combinatorial oxa-
diazole library. Bioactive molecules were discovered with
minimal experimental effort. The results demonstrate that
visualization of ‘chemical space’ provides an intuitive ap-
proach to molecular design and virtual screening in drug
discovery, even in the absence of a three-dimensional re-
ceptor structure.
Visualization of chemical data can help understand the
structure of compound distributions in chemical space and
guide molecular design experiments.[1–5] Commonly applied
visualization techniques in chemistry are principal compo-
nent analysis (PCA)[6] and self-organizing maps (SOMs, Ko-
honen networks).[7–9] Both methods have proven their value
for visualization of compound libraries and virtual screen-
ing. Still, they suffer from several drawbacks. For example, a
SOM’s quality to separate data depends on the chosen
map size, i.e. the number of ‘neurons’ (local clusters, Voro-
noi fields), and determining the actual quality of a comput-
ed SOM projection is nontrivial. A perceived disadvantage
is that SOMs lack immediate interpretability due to nonlin-
ear projection. While for low-dimensional data linear pro-
jection by PCA seems to be preferable, SOMs have shown
to produce more robust projections of high-dimensional
data.[10]
create some kind of order in data space. Once activity is-
lands are identified in the visualization, the compounds
that form such local clusters can be extracted and subject-
ed to biochemical tests.[12–15]
In the first application, we used a three-step technique
consisting of (i) data projection, (ii) visualization, and (iii)
clustering with the aim to identify new inhibitors of Helico-
bacter pylori (H. pylori) protease HtrA, a potential drug
target for combating the epithelial infection by human
pathogens and pathogen-mediated cancer.[16,17] In the
second application we aimed at target prediction for li-
gands obtained by combinatorial chemistry.
Prior to prospective screening, we assessed the reliability
and accuracy of the approach in a retrospective virtual
screening exercise (Table 1). We compiled a set of 11,244
bioactive drugs and lead structures from the literature
(COBRA collection v10.3),[18] and represented each com-
pound by two types of numeric features (‘descriptors’):
topological (two-dimensional, graph-based) pharmaco-
phore feature pairs (150-dimensional CATS descriptor),[19,20]
and three-dimensional (conformation-dependent) pseudo-
receptor models (90-dimensional PRPS descriptor).[21,22] We
defined potential activity islands by computing the convex
hulls for local groups of active compounds (for details, see
the supporting information). Averaged over ten targets,
CATS yielded a receiver operator characteristic (ROC) area
under the curve (AUC) of 0.61 (maximum=1), for PRPS we
obtained 0.69. Both methods demonstrated their ability to
[a] A. Klenner, V. Hꢀhnke, T. Geppert, P. Schneider, H. Zettl, S. Haller,
T. Rodrigues, F. Reisen, G. Schneider
Here, we present a method for visualizing and interpret-
ing high-dimensional chemical data, which is complemen-
tary to SOM and PCA projection and overcomes some of
their disadvantages and limitations. The projection is based
on stochastic proximity embedding (SPE).[11] SPE embeds
data in a low-dimensional space in such a way that pair-
wise distances between compounds are preserved. As a
consequence, patterns in the original high-dimensional
data distribution become accessible to visual inspection.
Making such patterns visible supports our intuitive interpre-
tation how a molecular representation might distinguish
between sets of compounds (e.g., active vs inactive) and
ETH, Department of Chemistry and Applied Biosciences
Wolfgang-Pauli-Str. 10, CH-8093 Zurich, Switzerland
*e-mail: gisbert.schneider@pharma.ethz.ch
[b] B. Hoy, S. Wessler
Paris-Lodron University, Department of Molecular Biology,
Division of Microbiology
Billroth Str. 11, A-5020 Salzburg, Austria
[c] A. M. Schaible, O. Werz
University of Jena, Institute of Pharmacy
Philosophenweg 14, D-07743 Jena, Germany
Supporting Information for this article is available on the WWW
under http://dx.doi.org/10.1002/minf.201100147
Mol. Inf. 2012, 31, 21 – 26
ꢁ 2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
21

Communication
A. Klenner et al.
Table 1. Results of retrospective virtual screening given as average
receiver-operator characteristic area under curve[48] (ROC-AUCꢀ
stddev.) computed from N similarity searches per target (N: number
of active reference compounds), and Matthews’ correlation coeffi-
cient[49] (mcc; interval [ꢁ1;1]). COX: cyclooxygenase, PPAR: peroxi-
some proliferator activated receptor.
Target
N
2D CATS descriptor 3D PRPS descriptor
ROC-AUC
mcc
ROC-AUC
mcc
COX-1
COX-2
5-lipoxygenase
(5-LO)
69 0.50ꢀ0.04
171 0.62ꢀ0.04
40 0.60ꢀ0.05
0.21
0.26
0.15
0.50ꢀ0.14
0.76ꢀ0.11
0.53ꢀ0.07
0.23
0.61
0.33
Thrombin
Trypsin
Factor Xa
PPARa
64 0.60ꢀ0.05
28 0.71ꢀ0.06
261 0.67ꢀ0.08
27 0.65ꢀ0.08
61 0.63ꢀ0.07
53 0.50ꢀ0.24
0.20
0.12
0.33
0.26
0.27
0.34
0.66ꢀ0.10
0.73ꢀ0.15
0.66ꢀ0.11
0.71ꢀ0.09
0.59ꢀ0.14
0.78ꢀ0.14
0.28
0.20
0.35
0.31
0.20
0.43
PPARg
Carbonic anhy-
drase II (CA II)
Dihydrofolate re-
ductase (DHFR)
61 0.66ꢀ0.09
0.23
0.93ꢀ0.02
0.72
Scheme 1. Reference HtrA inhibitor (1), and hits 2 and 3 found by
virtual screening.
retrieve actives from a set of drug-like decoys, while allow-
ing for scaffold-hops to occur. As ROC-AUC values were
higher for PRPS, we chose this molecular representation for
the prospective studies.
roDock[26] was used to rank the 228 compounds according
to their steric fit into the putative binding site of a compa-
rative protein model (‘homology model’) of H. pylori HtrA.
From the set of 100 top-ranking compounds, we ordered
18 diverse candidates from the supplier, and probed for
HtrA inhibition (Table S1). We emphasized structurally unex-
plored chemotypes so that potential scaffold-hops could
be explored. In vitro activity determination confirmed refer-
ence compound 1, and revealed compounds 2 and 3 as
dose-dependent inhibitors of E-cadherin cleavage by H.
pylori HtrA (Figure S2A). It is of note that the inhibitory ac-
tivity of compound 2 exceeds the activity level of the most
potent known HtrA inhibitor (HHI, IC50ffi25 mM) by approxi-
mately 50%, suggesting a preliminary IC50 value in the low
micromolar range (Figure S2B, S2C). We wish to point out
that Western blot quantification results in a crude estima-
tion of protease activity only. A more reliable direct binding
assay to monitor E-cadherin cleavage by HtrA is currently
under development in our laboratory. Compound 2 repre-
sents a tool compound for this purpose. It does not qualify
as a lead structure with regard to its relatively large molec-
ular weight (506 Da), many heteroatoms and a potential
Michael acceptor group with the nucleophilic attack at the
carbon next to the furanyl moiety. Of note, the inhibitory
effect of compound 2 on HtrA is unaffected by the pres-
ence of dithiothreitol (DTT), which points to structural in-
tegrity and a non-covalent mode of action (Figure 2).[27]
Future virtual screens will have to find scaffold replace-
ments aiming at smaller, non-reactive and more potent
HtrA inhibitors.
A first practical test of the virtual screening method was
to retrieve potential inhibitors of H. pylori protease HtrA.
This extracellular serine protease cleaves E-cadherin on
colonized gastric epithelial host cells and thereby enables
bacterial translocation across gastric epithelia.[23,24] The
Specs compound collection (as of January 2010; Specs,
Delft, The Netherlands) with a total of 127,753 screening
compounds (median molecular weight=381 Da) was repre-
sented by our PRPS pseudoreceptor descriptors. For refer-
ence compound 1 (Scheme 1), a moderate HtrA inhibitor
that was found by us in a previous structure-based virtual
screening study,[24] we computed 58 plausible conforma-
tions using a stochastic conformer generator.[25,26] Then, SPE
was used to project the complete compound cloud to
three-dimensions, and convex hulls were computed for po-
tential activity islands containing the conformations of HtrA
inhibitor 1.
Neighborhood analysis revealed that for any k=100
nearest neighbors in the three-dimensional projection, 90%
are also nearest neighbors in the original 90-dimensional
pseudoreceptor space, which demonstrates that global
neighborhood is largely preserved in the projected space
and it is legitimate to use such a visualization for com-
pound selection. The overall projection error was 11%. For
the volumes containing conformations of compound 1 it
was slightly lower (10%). We consider these error values as
permissive for visual data analysis and focused on the larg-
est potential activity island from which we retrieved 228
potential screening compounds that were grouped togeth-
er within the convex hull (Figure 1A, 1B). The software Neu-
As a second application of our convex hull algorithm, we
predicted potential targets for compounds that were as-
sembled by virtual combinatorial chemistry.[28] We generat-
22
www.molinf.com
ꢁ 2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2012, 31, 21 – 26

From Virtual Screening to Bioactive Compounds by Visualizing and Clustering of Chemical Space
Figure 1. Projected pseudoreceptor space containing 127,753 screening compounds and the conformations of reference compound 1. (a)
Complete chemical space with color scheme indicating the local projection error (blue: 0%, red: 58%). (b) Part of the projected space con-
taining the largest cluster of conformations of compound 1 (green spheres, outliers in red). 228 putative actives (orange) were found
inside the convex hull (lines). Other compounds are represented as black dots. For clarity, only 10% of the data are shown. (c) Projected
COBRA compounds (light grey dots) in pseudoreceptor space together with 464 oxadiazoles (green dots). Known 5-LO ligands are colored
in red. Cluster boundaries are shown as lines.
ed a virtual library of 464 oxadiazoles using commercially
available building blocks. Each virtual product was repre-
sented by its pseudoreceptor model and projected in a
space defined by a collection of 11,244 known drugs and
lead compounds with annotated targets (COBRA v10.3).[18]
Within the convex hulls spanned by a total of five oxadia-
zole clusters (Figure 1C), we found a prevalence of known
ligands of peroxisome proliferator activated receptors
(PPARs), 5-lipoxygenase (5-LO) and various G-protein cou-
pled receptors (GPCRs) (Figure S3). By analogy reasoning,
partially similar biological activity can be expected.[29–31] As
none of the virtual oxadiazoles contained an acidic head
group, which is an essential structural feature of most of
the known PPAR agonists,[32,33] we decided to investigate
only for 5-LO inhibition for the sake of this proof-of-con-
cept study. We chose compound
4
for synthesis
Mol. Inf. 2012, 31, 21 – 26
ꢁ 2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
23

Communication
A. Klenner et al.
with molecular pharmacophore representations are suited
for rapid target profile prediction, which might be useful
for off-target analysis and drug repurposing studies in phar-
maceutical research. It should be noted that any data pro-
jection results in a loss of information, and working in the
original, often high-dimensional descriptor space is
common practice for virtual screening. Our results show
that appropriate molecular representations and nonlinear
projection techniques result in low projection errors and
thus provide a means for visualization of chemical space,
which may serve as a supporting visual aid in virtual
screening campaigns.[36] Specifically, the low-dimensional
convex hulls may be regarded as conceptually similar to
visualizing the ‘applicability domain’ of pharmacophore
models.[37] In addition, the use of ‘fuzzy’ (permissive) phar-
macophore points like our pseudoreceptor approach could
help retrieve new chemotypes with a desired activity pro-
file. For hit-to-lead optimization, however, less abstract mo-
lecular descriptors (e.g. substructure fingerprints) might be
favorable.[38]
Figure 2. Inhibitory activity of compound 2 (100 mM) in the ab-
sence (left) and presence (right) of DTT (100 ng E-cadherin; 100 ng
HtrA in 50 mM Hepes buffer, pH 7.4; ꢀ1 mM DTT; incubation for
16 h at 378C).
(Scheme 2), because it represents the oxadiazole represen-
tative located closest to the centroid of the known 5-LO in-
hibitors.
Experimental
Experimental results from the 5-LO inhibition assay sug-
gest that compound 4 is a partial inhibitor of 5-LO. Its in-
hibitory effect was significant at concentrations ꢂ10 mM
(Figure S4). Neither the related 12- and 15-LOs nor cycloox-
ygenases (COX)-1 and -2 were inhibited (not shown), which
is in perfect agreement with the computer-based predic-
tion. At this time the mode of action of compound 4 is un-
known. Apparently compound 4 exhibits an activity profile
similar to 5-LO modulators that do not interfere with the
catalytic domain directly, which we conclude from the ob-
servation that inhibition leveled at approximately 30%
even at raised concentrations (Figure S4). For example, hy-
perforin was shown to modulate enzyme activity by inter-
fering with the C2-like domain of 5-LO.[34] We wish to point
out that high compound concentrations can result in ag-
gregation and measurement artifacts.[35] While we cannot
exclude aggregation of compound 4 at high concentra-
tions, the inhibitory effect appears to be dose-dependent
at low concentrations but exhibits a saturation effect. Fur-
ther investigation is needed to explain this experimental
observation.
Computational. Preparation of compound libraries was
done with MOE (Molecular Operating Environment, v.2010,
The Chemical Computing Group, Montreal, Canada) using
the ‘wash’ function with default settings. CORINA v3.46
(Molecular Networks, Erlangen, Germany) was used to gen-
erate single three-dimensional conformations. CATS[19] and
PRPS[21] descriptors were computed as described.[39,40]
A
comparative protein model (‘homology model’) of H. pylori
HtrA was generated using MOE Homology (v2007.09)[41]
taking Escherichia coli HtrA (PDB[42] ID: 3cs0,[43] 3 ꢂ resolu-
tion, 41% identity to H. pylori HtrA) as template, as de-
scribed.[44]
HtrA inhibition assay. Cloning, expression and purification
of H. pylori HtrA was performed as described previously.[45]
The ordered test compounds were dissolved in DMSO and
diluted to stock concentration (2 mM). 0.5 mg HtrA was in-
cubated with 100 mM of the respective compound and
0.1 mg E-cadherin/Fc-Chimera (R&D Systems) in 50 mM 4-(2-
hydroxyethyl)-1-piperazineethane sulfonic acid (HEPES)
buffer (pH 7.4) for two hours at 378C. The reaction was
stopped by boiling for five minutes and analyzed by West-
ern-blotting using anti-E-cadherin (Santa Cruz Biotech-
The results of our study corroborate the pseudoreceptor
approach for ligand-based target prediction. We demon-
strate that data projection and visualization in combination
Scheme 2. Synthesis of oxadiazole 4. DCM: dichlormethane, DME: 1,2-dimethoxyethane.
24
www.molinf.com
ꢁ 2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2012, 31, 21 – 26

From Virtual Screening to Bioactive Compounds by Visualizing and Clustering of Chemical Space
nology) or anti-HtrA antibodies. A film was exposed to the
ECL/HRP chemo-luminescence reaction and scanned, or
data were acquired directly by a ChemiDoc XRS system
(BioRad). Background noise filtering by a rolling-ball algo-
rithm and the measurement of brightness densities was
performed using ImageJ (version1.41o).[46]
H), 7.69 (1H, td, J=8.0; 1.6 Hz, Ar-H); 7.79 (1H, td, J=8.0;
1.6 Hz, Ar-H); 7.83 (1H, dd, J=8.0; 1.6 Hz, Ar-H); 7.94 (2H,
m, Ar-H); 8.08 (2H, m, Ar-H); 8.23 (1H, dd, J=8.0; 1.2 Hz, Ar-
H).
Acknowledgements
Determination of 5-LO product formation in cell-based
assays. Human neutrophils were freshly isolated from leu-
kocyte concentrates obtained at the Institute for Transfu-
sion Medicine, University Hospital Jena (Germany) as de-
scribed.[47] In brief, venous blood was taken from healthy
adult donors that did not take any medication for at least
7 days, and leukocyte concentrates were prepared by cen-
trifugation (4000ꢃg, 20 min, 208C). Neutrophils were im-
mediately isolated by dextran sedimentation and centrifu-
gation on Nycoprep cushions (PAA Laboratories, Linz, Aus-
tria) and hypotonic lysis of erythrocytes was performed.
Cells were finally re-suspended in PBS pH 7.4 containing
1 mg/mL glucose and 1 mM CaCl2 (PGC buffer) (purity>
96–97%). For assays of intact cells, 5ꢃ106 freshly isolated
neutrophils were re-suspended in 1 mL PGC buffer. After
pre-incubation with the test compounds for 10 min at
378C, 5-LO product formation was started by addition of
2.5 mM A23187 and exogenous AA (20 mM). After 10 min at
378C, the reaction was stopped with 1 mL of methanol and
30 mL of 1 N HCl, 200 ng prostaglandin B1 and 500 mL of
PBS were added. Formed 5-LO metabolites were extracted
and analyzed by HPLC, as described.[47] 5-LO product forma-
tion is expressed as ng of 5-LO products per 106 cells,
which includes LTB4 and its all-trans isomers, (5S,12S)-di-hy-
droxy-6,10-trans-8,14-cis-eicosatetraenoic acid ((5S,12S)-
DiHETE), and (5S)-hydro(pero)xy-6-trans-8,11,14-cis-eicosate-
traenoic acid (5-H(p)ETE). Cysteinyl LTsC4, D4 and E4 were
not detected. Oxidation products of LTB4 were not deter-
mined.
The OPO-Foundation Zꢁrich and the Deutsche Forschungsge-
meinschaft (FOR1406, TP4) are thanked for financial sup-
port. The Chemical Computing Group Inc., Montreal,
Canada, provided a research license of MOE.
References
[1] S. J. Campbell, A. Gaulton, J. Marshall, D. Bichko, S. Martin, C.
Brouwer, L. Harland, Drug Discov. Today 2010, 15, 3–15.
[2] M. Wawer, E. Lounkine, A. M. Wassermann, J. Bajorath, Drug
Discov. Today 2010, 15, 630–639.
[3] D. M. Maniyar, I. T. Nabney, B. S. Williams, A. Sewing, J. Chem.
Inf. Model. 2006, 46, 1806–1818.
[4] G. Schneider, Curr. Med. Chem. 2002, 9, 2095–2101.
[5] T. I. Oprea, J. Gottfries, J. Comb. Chem. 2001, 3, 157–166.
[6] K. Pearson, Philos. Mag. 1901, 2, 559–572.
[7] T. Kohonen, Biol. Cybern. 1982, 43, 59–69.
[8] J. Gasteiger, J. Zupan, Neural Networks in Chemistry and Drug
Design, Wiley-VCH, Weinheim, 1999.
[9] P. Schneider, Y. Tanrikulu, G. Schneider, Curr. Med. Chem. 2009,
16, 258–266.
[10] M. Schmuker, G. Schneider, Proc. Natl. Acad. Sci. USA 2007,
104, 20285–20289.
[11] D. K. Agrafiotis, H. Xu, F. Zhu, D. Bandyopadhyay, P. Liu, Mol.
Inf. 2010, 29, 758–770.
[12] R. Guha, J. H. Van Drie, Chem. Inf. Model. 2008, 48, 646–658.
[13] G. Schneider, M. Hartenfeller, M. Reutlinger, Y. Tanrikulu, E. Pro-
schak, P. Schneider, Trends Biotechnol. 2009, 27, 18–26.
[14] M. T. Sisay, L. Peltason, J. Bajorath, J. Chem. Inf. Model. 2009,
49, 2179–2189.
Synthesis of compound 4 ((2-(3-(4-fluorophenyl)-1,2,4-oxa-
diazol-5-yl)benzyl benzoate). 200 mg of educt E1 (Z)-4-
fluoro-N’-hydroxybenzimidamide (154.1 Da) and 392 mg of
educt E2 2-(chlorocarbonyl)benzyl benzoate (274.7 Da)
were dissolved in 200 mL dichloromethane. The solution
was stirred for 6 h at room temperature to facilitate inter-
mediate imine formation. The intermediate product was re-
crystallized in n-hexane/ethyl acetate and dissolved in
200 mL 1,2-dimethoxyethane. The solution was heated to
1608C for 6 hours. Final product was obtained by flash
column chromatography (Biotage Isolera; n-hexane/EtOAc;
yield: 112 mg=28% over two steps). Analytics (Figure S4):
purity (area normalization)>99%, m/z=375 (Shimadzu LC-
MS2020; HPLC: H2O+0.1% trifluorocetic acid (TFA)/30–
[15] S. Renner, M. Hechenberger, T. Noeske, A. Bçcker, C. Jatzke, M.
Schmuker, C. G. Parsons, T. Weil, G. Schneider, Angew. Chem.
Int. Ed. 2007, 46, 5336–5339.
[16] D. Zurawa-Janicka, J. Skorko-Glonek, B. Lipinska, Expert Opin.
Ther. Targets 2010, 14, 665–679.
[17] G. Haas, G. Karaali, K. Ebermayer, W. G. Metzger, S. Lamer, U.
Zimny-Arndt, S. Diescher, U. B. Goebel, K. Vogt, A. B. Roznow-
ski, B. J. Wiedenmann, T. F. Meyer, T. Aebischer, P. R. Jungblut,
Proteomics 2002, 2, 313–324.
[18] P. Schneider, G. Schneider, QSAR Comb. Sci. 2003, 22, 713–
718.
[19] G. Schneider, W. Neidhart, T. Giller, G. Schmid, Angew. Chem.
Int. Ed. 1999, 38, 2894–2896.
[20] U. Fechner, G. Schneider, QSAR Comb. Sci. 2004, 23, 19–22.
[21] Y. Tanrikulu, E. Proschak, T. Werner, T. Geppert, N. Todoroff, A.
Klenner, T. Kottke, K. Sander, E. Schneider, R. Seifert, H. Stark, T.
Clark, G. Schneider, ChemMedChem 2009, 4, 820–827.
[22] Y. Tanrikulu, G. Schneider, Nat. Rev. Drug Discov. 2008, 7, 667–
677.
[23] M. Lçwer, T. Geppert, P. Schneider, B. Hoy, S. Wessler, G.
Schneider, PLoS One 2011, 6, e17986.
[24] B. Hoy, M. Lçwer, C. Weydig, G. Carra, N. Tegtmeyer, T. Gep-
pert, P. Schrçder, N. Sewald, S. Backert, G. Schneider, S. Wess-
ler, EMBO Rep. 2010, 11, 798–804.
1
95% acetonitrile+0.1% TFA, RP18, 250 nm, ESI+); H-NMR
[Bruker Avance 400; 400.13 MHz, DMSO-d6, proton–proton
1
coupling constants (J) are given in Hertz (Hz), H NMR peak
multiplicity is indicated by s (singlet), d (doublet), dd (dou-
blet of doublet), td (triplet of doublets), tt (triplet of trip-
lets), m (unresolved multiplet)]: 5.85 (2H, s, CH2); 7.43 (2H,
m, Ar-H); 7.49 (2H, m Ar-H); 7.63 (1H, tt, J=8.0; 1.6 Hz, Ar-
Mol. Inf. 2012, 31, 21 – 26
ꢁ 2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.molinf.com
25

Communication
[25] H. Xu, S. Izrailev, D. K. Agrafiotis, J. Chem. Inf. Comput. Sci.
2003, 43, 1186–1191.
[26] A. Klenner, M. Weisel, F. Reisen, E. Proschak, G. Schneider, Mol.
Inf. 2010, 29, 75–86.
[27] a) S. Ludewig, M. Kossner, M. Schiller, K. Baumann, T. Schir-
meister, Curr. Topics Med. Chem. 2010, 10, 368–382; b) J. E.
Blanchard, N. H. Elowe, C. Huitema, P. D. Fortin, J. D. Cechetto,
L. D. Eltis, E. D. Brown, Chem. Biol. 2004, 11, 1445–1453.
[28] F. H. Reisen, G. Schneider, E. Proschak, J. Chem. Inf. Model.
2009, 49, 6–12.
A. Klenner et al.
Curr. Comput. Aided Drug Des. 2011, 7, 1–9; c) R. Guha, K. Gil-
bert, G. Fox, M. Pierce, D. Wild, H. Yuan, Curr. Comput. Aided
Drug Des. 2010, 6, 50–67.
[38] A. Klenner, M. Hartenfeller, P. Schneider, G. Schneider, Drug
Discov. Today Technol. 2010, 7, e237–e244.
[39] U. Fechner, L. Franke, S. Renner, P. Schneider, G. Schneider, J.
Comput. Aided Mol. Des. 2003, 17, 687–698.
[40] C. B. Rçdl, Y. Tanrikulu, J. M. Wisniewska, E. Proschak, G.
Schneider, D. Steinhilber, B. Hofmann, ChemMedChem 2011, 6,
1001–1005.
[29] A. Bender, R. C. Glen, Org. Biomol. Chem. 2004, 2, 3204–3218.
[30] M. J. Keiser, V. Setola, J. J. Irwin, C. Laggner, A. I. Abbas, S. J.
Hufeisen, N. H. Jensen, M. B. Kuijer, R. C. Matos, T. B. Tran, R.
Whaley, R. A. Glennon, J. Hert, K. L. Thomas, D. D. Edwards,
B. K. Shoichet, B. L. Roth, Nature 2009, 462, 175–181.
[31] R. Steri, P. Schneider, A. Klenner, M. Rupp, J. M. Kriegl, M. Schu-
bert-Zsilavecs, G. Schneider, Mol. Inf. 2010, 29, 287–292.
[32] V. Zoete, A. Grosdidier, O. Michielin, Biochim. Biophys. Acta
2007, 1771, 915–925.
[33] Y. H. Peng, M. S. Coumar, J. S. Leou, J. S. Wu, H. Y. Shiao, C. H.
Lin, W. H. Lin, T. W. Lien, X. Chen, J. T. Hsu, Y. S. Chao, C. F.
Huang, P. C. Lyu, H. P. Hsieh, S. Y. Wu, ChemMedChem 2010, 5,
1707–1716.
[34] C. Feisst, C. Pergola, M. Rakonjac, A. Rossi, A. Koeberle, G.
Dodt, M. Hoffmann, C. Hoernig, L. Fischer, D. Steinhilber, L.
Franke, G. Schneider, O. Rꢄdmark, L. Sautebin, O. Werz, Cell.
Mol. Life Sci. 2009, 66, 2759–2771.
[35] a) B. K. Shoichet, Drug Discov. Today 2006, 11, 607–615; b) J.
Seidler, S. L. McGovern, T. N. Doman, B. K. Shoichet, J. Med.
Chem. 2003, 46, 4477–4486.
[41] M. A. Marti-Renom, A. C. Stuart, A. Fiser, R. Sꢅnchez, F. Melo, A.
Sali, Ann. Rev. Biophys. Biomol. Struct. 2000, 29, 291–325.
[42] H. M. Berman, J. Westbrook, Z. Feng, G. Gilliland, T. N. Bhat, H.
Weissig, I. N. Shindyalov, P. E. Bourne, Nucleic Acids Res. 2000,
28, 235–242.
[43] T. Krojer, J. Sawa, E. Schꢀfer, H. R. Saibil, M. Ehrmann, T. Clau-
sen, Nature 2008, 453, 885–890.
[44] M. Lçwer, Virtuelles Screening nach Inhibitoren der Protease
HtrA aus Helicobacter pylori, PhD Thesis, Goethe-University
Frankfurt a.M., Germany, 2009.
[45] M. Lçwer, C. Weydig, D. Metzler, A. Reuter, A. Starzinski-
Powitz, S. Wessler, G. Schneider, PLoS One 2008, 3, e3510.
[46] M. D. Abramoff, P. J. Magalhaes, S. J. Ram, Biophotonics Inter-
nat. 2004, 11, 36–42.
[47] C. Pergola, G. Dodt, A. Rossi, E. Neunhoeffer, B. Lawrenz, H.
Northoff, B. Samuelsson, O. Rꢄdmark, L. Sautebin, O. Werz,
Proc. Natl. Acad. Sci USA 2008, 105, 19881–19886.
[48] K. H. Zou, A. J. O’Malley, L. Mauri, Circulation 2007, 115, 654–
657.
[49] B. W. Matthews, Biochim. Biophys. Acta 1975, 405, 442–451.
[36] M. Reutlinger, W. Guba, R. E. Martin, A. I. Alanine, T. Hoffmann,
A. Klenner, J. A. Hiss, P. Schneider, G. Schneider, Angew. Chem.
Int. Ed. 2011, doi: 10.1002/anie.201105156, in press.
[37] a) J. Jaworska, N. Nikolova-Jeliazkova, T. Aldenberg, Altern. Lab.
Anim. 2005, 33, 445–459; b) K. Funatsu, T. Miyao, M. Arakawa,
Received: September 23, 2011
Accepted: November 4, 2011
26
www.molinf.com
ꢁ 2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Mol. Inf. 2012, 31, 21 – 26