WISDOM: A Grid-Enabled Drug Discovery Initiative against Malaria - Grid Computing: Infrastructure, Service, and Applications - page 363

validated at the University of Modena in Giulio Rastelli team [15]. The
workl ow, based on the Amber package, is able to automatically and efi -
ciently rei ne docking poses, which sometimes may not be accurate, and
rank the compounds based on more accurate scoring functions. The pro-
cedure called BEAR (Binding Estimation After Rei nement) requires as
input a pdb i le containing the structure of the protein and a mol2 i le
containing the coordinates of the docked ligands. The coordinates of the
docked ligand and target structure are merged to create the complex.
The topology i les are created using antechamber. The ligand atoms are
described with GAFF (general Amber force i eld) atom types and AM1-
BCC charges. In order to avoid the time-consuming procedure of charge
calculation, atomic charges of the ligand are read from the original mol2
i le and not computed during the procedure; this choice adds the advan-
tage that charge calculation of ligands can be performed only once, and
obtained mol2 i les can be used for many other target proteins. The ligand
charges are calculated using antechamber by means of a separate script.
Interestingly, the same set of AM1-BCC charges can also be exploited for
automated ligand docking and MD. Missing GAFF parameters for the
ligand are automatically assigned by parmcheck. The ligand, receptor, and
complex topologies are written using leap utility (Amber 9). Minimization,
MD, and i nal reminimization of the complexes are performed using sander
with a distance-dependent dielectric constant e = 4 r . For each of these
steps, the procedure enables the user to set ad hoc rei nement options
depending on the application. After rei nement of the complex, a pdb i le
is generated as output and the i nal coordinates of the ligand, receptor,
and complex are updated and used to compute the binding free energy
evaluation using Amber MM-PBSA and MM-GBSA. The free energy
results (D G MM , D G solv , and D G bind ) are written to an output i le and com-
pounds are ranked on the basis of their binding free energy [15].
This workl ow was implemented in the EGEE grid using the WISDOM
production environment designed for the large-scale docking experi-
ments to deploy the MD procedure on the grid. The wide CPU availability
of the EGEE grid allowed the submission of a large subset of best docking
complexes to the MD rei nement and rescoring procedure despite its high
computational demand.
The i rst successful deployment of the MD procedure was achieved in
December 2006 on fake data. At the preparation stage, all the required
input i les and Amber executables were stored on the storage element
(SE) of the grid. At the execution stage, jobs lasting for approximately 20
hours were submitted corresponding to 50 compound subsets. Each sub-
set with 50 compounds was submitted on one worker node along with
the Amber executables, target structure, and the main script controlling
the MD rei nement and rescoring operations. After the jobs were i nished
the result i les (structures and energy scores) were copied and stored on
the storage element.