A Method for Similarity-Based Grouping of Biological Data

Abstract

Abstract. Similarity-based grouping of data entries in one or more data sources is a task underlying many different data management tasks, such as, structuring search results, removal of redundancy in databases and data integration. Similarity-based grouping of data entries is not a trivial task in the context of life science data sources as the stored data is complex, highly correlated and represented at different levels of granularity. The contribution of this paper is two-fold. 1) We propose a method for similarity-based grouping and 2) we show results from test cases. As the main steps the method contains specification of grouping rules, pairwise grouping between entries, actual grouping of similar entries, and evaluation and analysis of the results. Often, different strategies can be used in the different steps. The method enables exploration of the influence of the choices and supports evaluation of the results with respect to given classifications. The grouping method is illustrated by test cases based on different strategies and classifications. The results show the complexity of the similarity-based grouping tasks and give deeper insights in the selected grouping tasks, the analyzed data source, and the influence of different strategies on the results.

A Tool for Evaluating Strategies for Grouping of Biological Data

“…In [15] a method that supports similarity-based grouping of biological data and that enables the development of grouping procedures was proposed. It covers the steps proposed for general clustering in [13] and duplicate detection in [16].…”

“…We use the test cases proposed in [15] and re-evaluate these using KitEGA. For a more detailed description of the test cases we refer to [15]. We also note that the overhead of storing the results of the running of the grouping procedures is minimal.…”

“…The similarity between GO terms can be used to compute a similarity between data entries that are annotated with these GO terms (e.g. [17,15]). …”

“…KitEGA is based on the method proposed in [15]. It allows evaluators to plug in their own algorithms related to the grouping strategies and the evaluation measures, as well as their own data sets, and provides support for the analysis of the grouping results.…”

“…It allows evaluators to plug in their own algorithms related to the grouping strategies and the evaluation measures, as well as their own data sets, and provides support for the analysis of the grouping results. The method from [15] and other background is discussed in section 2. We present KitEGA and its current prototype implementation in section 3.…”