A Tool for Evaluating Strategies for Grouping of Biological Data

During the last decade an enormous amount of biological data has been generated and techniques and tools to analyze this data have been developed. Many of these tools use some form of grouping and are used in, for instance, data integration, data cleaning, prediction of protein functionality, and correlation of genes based on microarray data. A number of aspects influence the quality of the grouping results: the data sources, the grouping attributes and the algorithms implementing the grouping procedure. Many methods exist, but it is often not clear which methods perform best for which grouping tasks. The study of the properties, and the evaluation and the comparison of the different aspects that influence the quality of the grouping results, would give us valuable insight in how the grouping procedures could be used in the best way. It would also lead to recommendations on how to improve the current procedures and develop new procedures. To be able to perform such studies and evaluations we need environments that allow us to compare and evaluate different grouping strategies. In this paper we present a framework, KitEGA1, for such an environment, and present its current prototype implementation. We illustrate its use by comparing grouping strategies for classifying proteins regarding biological function and isozymes.