Speaker
Mr
Hurng-Chun Lee
(ASGC, Taiwan)
Description
Recent studies have suggested that the high pathogenic avian
flu H5N1 virus has the
potential of developing drug resistance and of acquiring the
ability of
human-to-human transmission. To enable biologists a better
response to the threat,
the second EGEE biomedical data challenge battling avian flu
was set to screen
300,000 compounds against 8 predicted mutations of the
Influenza A Neuraminidase for
analyzing the efficiency of the known drugs and for
searching new drugs. In April and
May 2006, we succeeded to mobilize over 2,000 CPUs in the
EGEE Grid infrastructure,
demonstrating that the high-throughput screening (HTS) of
drug analysis can be
efficiently reproduced on the Grid using the WISDOM platform
previously developed for
the Malaria data challenge in last Summer. The 6-weeks
activity has covered over 100
CPU years of CPU power required for the virtual screening
process and has produced
about 600 Gigabytes of docking results for further analysis.
Current computing model of the Grid-enabled HTS adopts a
coordinative way of
execution in order to gain the docking throughput; however,
to bring biologists a
real end-user application for their daily research, the
application usability needs
to be improved taking into account the realistic usage
patterns. For example, the
preparation and deployment effort needed for starting the
data challenge will not be
appreciated by the users who frequently repeat the virtual
screening for testing
their libraries and docking parameters. The batch mode HTS
is also not feasible for
interactive analysis which can save biologists’ time by
allowing them to start
analyzing partial results on the fly instead of dealing with
a huge amount of output
at the end. In addition, biologists prefer a graphical user
interface to configure
domain-specific parameters.
To improve the usability, we first introduced a light-weight
framework called DIANE
to enable the interactive analysis of the Grid-enabled
virtual screening application.
DIANE was originally developed for handling the distributed
applications within a
Master-Worker model. It provides an overlay system on top of
the Grid system, in
which the pull-mode scheduling and failure recovery
mechanisms are implemented based
on the CORBA protocol. On the other hand, the DIANE
framework hides the details of
the job operations on the Grid so that application
developers can concentrate on the
implementation of application logic, and end users benefit
from the simplified job
descriptions containing only intuitive and application
specific parameters. The
stability and efficiency of DIANE has been tested by taking
a significant part of the
avian flu data challenge.
Following the successful avian flu data challenge, the
Academia Sinica Grid Computing
Centre (ASGC) in Taiwan is in charge of developing a
user-friendly docking
environment leveraging on the DIANE framework. Building on
top of the DIANE
command-line interface, we have customized a web-based
interface for biology
end-users. Through the web interface, users can quickly set
a filter on compound
libraries, configure docking parameters, start-up and
monitor their virtual screening
activities on the EGEE production environment. During the
job execution, the
completed dockings are scored based on the binding energy.
The docking complexes are
available as soon as the results are produced; therefore
biologists can progress on
for further analysis without being blocked until their jobs
are finished. A
visualization interface of complex structures also aids
biologists in analysis.
Primary author
Mr
Li-Yung Ho
(ASGC, Taiwan)
Co-authors
Mr
Hurng-Chun Lee
(ASGC, Taiwan)
Dr
Ying-Ta Wu
(Academia Sinica, Taiwan)
