The temptation of high-throughput docking
Possible strategies and the development of required tools
Christof Gerlach, Peter Block, Gerhard Klebe
Department of Pharmaceutical Chemistry, Marbacher Weg 6, 35032 Marburg (Germany)
Target
Background
Recent improvements in both software and hardware principally
As a test example, we used the binding of Fusicoccin, a fungal
allows to use docking in high-throughput mode as a tool for
phytotoxin stabilizing the interaction between the C-terminus of
virtual screening without prior application of sophisticated
the plant plasma membrane H+-ATPase and 14-3-3 proteins. This
pharmacophore filters. Due to computational demands, ligand
stabilization leads to permanent activation of the proton pump
docking is usually applied as final step in virtual screening.
resulting in wilting of the plant.
As it is no longer the methodological bottleneck, it can now be
Recently, the crystal structure of the
used to generate ensembles of binding confomers which have to
ternary complex between a plant 14-3-3
be postprocessed in terms of sophisticated pharmacophore
protein, Fusicoccin, and a pentapeptide
models and robust scoring schemes. This puts increasing
representing the C-terminus of the H+-
demand on the reliability of the latter tools and affords efficient
ATPase has been solved[1]. We selected
protocols for analyzing protein-ligand interaction geometries to
the Fusiccocin binding site as target for
filter vast amounts of docking solutions for various molecules to
our high-throughput docking attempt to
retrieve reliably the most promising candidates in automated
screen for potential new ligands stabilizing
the protein-protein interaction.
C-terminus of the H+-ATPase (wheat), and Fusicoccin (FC).
We used a subset of the Maybridge screening database as
Benchmark
FlexX
toxin Fusicoccin (FC).
Ternary complex of 14-3-3 protein (blue), a pentapeptid representing the
fashion.
2.0[2]
The structure of the fungal
dataset for our high-throughput docking. We filtered for molecules
was used as docking tool and tested on i386 and
amd64 architectures. The benchmark test was performed to
compare the docking performance between Intel Pentium 2.8
Unity
with a molecular weight between 200 and 450 and rejected all
molecules with more than 6 rotatable bonds. The subset is
composed of 42640 molecules.
GHz, AMD AthlonXP 2400+ 2.0 GHz, and AMD Opteron 1.8 GHz
OpenPBS
machines, each equipped with 1GB RAM.
Distributor
We selected conditions to avoid any inefficiencies due to
networking or hard disk performance. All 42640 molecules were
docked on each platform.
Seeker
Cluster
Cluster
DDB
Unity 3D Query
molecules / minute
Pentium4
Athlon XP
5.8
7.1
The Seeker is the Graphical User Interface (GUI) controlling the
A comparative flexible 3D Unity[4] search extracted 86 compounds
whole docking process via the Distributor in combination with the
out of the 42640 molecules of the prepared Maybridge database.
queuing system OpenPBS. The docking results are stored in the
The pharmacophore criteria are fulfilled by all molecules.
by the 64bit AMD Opteron by nearly a factor 2. Furthermore, the rather
Oracle-based Docking Database (DDB)[3]. The Seeker enables
Analysing the proposed rough geometries from Unity within the
outdated AMD AthlonXP 2400+ shows 20% speed advantage over the Intel
us
binding site shows mostly clashes with the aminoacids of the 14-3-3
Opteron
11.3
Although the 32bit Intel Pentium4 is clocked faster by 50%, it is outperformed
to
analyze
each
docking
placement
and
interaction
Pentium4 2.8 GHz CPU. We clearly recommend the AMD Opteron platform to
perform high-throughput docking with FlexX.
protein.
conveniently.
Tyr-137
Filter
R
R
Tyr-134
Asp-133
R
OH
R
O
Lys-129
FlexX generated 50 placements for each of the 42640 molecules
To rank the filtered moleHO
O
+
NH3
in the dataset. To retrieve promising candidates we filtered the
vast amount of around 2 million docking solutions with the
Val-5
Docking Database (DDB). The database enables us to create
R
NH2
O
Lys-56
O
N
A
HO
H
Ile-226
S
Met-130
R
O
Asp-222
O
bond acceptor.
scoring step.
shows
predictive
im-
power
the
original
Drug-
corePDB and gives valuable
details about specific interactions towards moieties of
H2N
O
R
interest,
which
can
Asn-49
intuitively visualized.
We translated these essential contacts into a DDB interaction
filter. Its application resulted in 154 candidates for the subsequent
Drug-
Ser-52
R
D
Ile-175
hydrophobic contact. Lys-129 and Tyr-134 are serving as
hydrogen bond donors, whereas Asp-222 represents a hydrogen
proved
R
Potential binders for the FC pocket are supposed to experience
following interactions: The C-terminal Val-5 must be covered by a
applied
scoreCSD, which was newly
over
R
we
DrugscoreCSD
R
filters to extract only those docking geometries in agreement with
the required interaction pattern.
cules,
developed in our group.
A
R
Scoring
Interaction pattern in the FC binding site. Important residues (Val-5, Lys-129,
Tyr-134, Asp-222) are displayed with intersecting interaction fields. On the
schematic ligand the desired interaction types are indicated as Acceptor,
Donor, Hydrophobic respectively.
be
DrugScoreCSD visualization of a docking
solution. Blue spheres are attractive and
red spheres repulsive interactions. The
size of the spheres represent the
contribution of each atom to the total
score.
The visual inspection of the 154 molecules confirmed that both the
filter setup and the proposed docking geometries are of very high
quality. In conclusion, high-throughput docking turns out to be a
reasonable alternative to the classical virtual screening approach.
References
[1] Wurtele M, Jelich-Ottmann C, Wittinghofer A, Oecking C, EMBO J, 2003, 22, 987.
[2] Claussen H, Gastreich M, Apelt V, Greene J, Hindle S, Lemmen C, CDDT, 2004, 1, 49.
[3] Rarey M, Kramer B, Lengauer T, Klebe G, J Mol Biol, 1996, 261, 470.
[4] Martin Y.C., J Med Chem 1992 35:2145
Acknowledgements
We thank BioSolveIT GmbH (St. Augustin, Germany) for providing the Flex* software platform, and
especially Volker Apelt, Frank Sonnenburg and Holger Claussen for technical support and fruitful
discussions. We also thank Hans Velec for his assistance with DrugscoreCSD.