Support for Cooperative Experiments in e-Science: From Scientific Work flows to Knowledge Sharing - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 138

not sufficient - we also require tools for deriving knowledge and validating models
(Miles et al. 2007a ). Support for the design of scientific workflows faces a number
of challenges:

￿ semantic support for annotating and searching workflows,

￿ reproducibility of results, configurations, and the runtime conditions of the

experiments,
interoperability between different workflows when they are used in one ￿

experiment.

Execution : most of the support provided by current SWMSs is dedicated to execu-
tion. This step of a scientific experiment is concerned with the execution of tasks
comprising the workflow, which can be entirely computational, but can also involve
interaction or manual steps that have to be performed by the user. Specific require-
ments from the applications include access to statistical toolkits as well as simple
access to parallel computation. Experiment execution support also involves the
delegation and automation of non-scientific and redundant tasks to the framework
such as the staging of software components constituting the experiment and the data
sets needed for the experiments, the search for the appropriate and available
computing resources, and monitoring of experiment progress (Mayer et al. 2006 ;
Olabarriaga et al. 2007 ). Interactive execution control is often required to allow
scientists to steer the execution path and tune component parameters, in particular
for the calibration phases of scientific experiments where the scientist is still experi-
menting with various parameter sets.

Analysis : this step of the scientific experimentation is the most delicate as it
re-quires a lot of scientific knowledge which is often difficult to model, and thus
calls for interaction with and among the scientists. It focuses on checking whether
the output of a workflow complies with the theory or expected results. During
execution, monitoring progress and steering execution between different paths are
basic human activities that control the experiment. There are several scenarios
which re- quire collaborative control between scientists. For instance, control of
the processes in a complex experiments requires analysis of results, which often
represents a joint effort by several scientists. Besides, a complex experiment con-
sists of more than one workflow and these workflows might be shared and modified
by geographically distributed scientists at the same time. As such, controlling an
entire complex experiment may require input from all these scientists (Humphrey
and Hamilton 2004 ) .

Dissemination : Traditionally, dissemination is achieved through scientific publi-
cation. It consists in making the resources and workflows themselves available for
use by others, providing specific metadata about the circumstances in which the
results were created (also known as provenance), as well as sharing the knowledge
associated with the proper use of resources, and providing abstracted versions of
successful workflows (Miles et al. 2007b ). Contemporary means of communicating
and sharing scientific results (i.e. by publishing papers) fall short of the requirements
of modern computational sciences, as such results do not easily lend themselves to