Cloud computing for fast prediction of chemical activity
Quantitative Structure-Activity Relationships (QSAR) is a method for creating models that can predict certain properties of compounds. It is of growing importance in the design of new drugs. The quantity of data now available for building models is increasing rapidly, which has the advantage that mo...
Saved in:
Published in | Future generation computer systems Vol. 29; no. 7; pp. 1860 - 1869 |
---|---|
Main Authors | , , , |
Format | Journal Article |
Language | English |
Published |
Elsevier B.V
01.09.2013
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | Quantitative Structure-Activity Relationships (QSAR) is a method for creating models that can predict certain properties of compounds. It is of growing importance in the design of new drugs. The quantity of data now available for building models is increasing rapidly, which has the advantage that more accurate models can be created, for a wider range of properties. However the disadvantage is that the amount of computation required for model building has also dramatically increased. Therefore, it became vital to find a way to accelerate this process. We have achieved this by exploiting parallelism in searching the QSAR model space for the best models. This paper shows how the cloud computing paradigm can be a good fit to this approach. It describes the design and implementation of a tool for exploring the model space that exploits our e-Science Central cloud platform. We report on the scalability achieved and the experiences gained when designing the solution. The acceleration and absolute performance achieved is much greater than for existing QSAR solutions, creating the potential for new, interesting research, and the exploitation of this approach to accelerate other types of applications.
► A design of a scalable QSAR system based on cloud and e-Science Central is proposed. ► System reaches limit of 200 nodes with over 90% of ideal speed-up. ► The key steps of migrating e-Science Central to the cloud are presented. ► Scalability, throughput and error rate of the system are studied. |
---|---|
ISSN: | 0167-739X 1872-7115 |
DOI: | 10.1016/j.future.2013.01.011 |