Computing Science. Cloud Computing for Chemical Activity Prediction

摘要

This paper describes how cloud computing has been used to reduce the time taken to generate chemical activity models from years to weeks. Chemists use Quantitative Structure-Activity Relationship (QSAR) models to predict the activity of molecules. Existing Discovery Bus software builds these models automatically from datasets containing known molecular activities, using a panel of experts algorithm. Newly available datasets offer the prospect of generating a large number of significantly better models, but the Discovery Bus would have taken over 5 years to compute them. Fortunately, we show that the panel of experts algorithm is well-matched to clouds. In the paper we describe the design of a scalable, Windows Azure based infrastructure for the panel of experts pattern. We present the results of a run in which up to 100 Azure nodes were used to generate results from the new datasets in 3 weeks.