Evaluating Dynamic Services in Bioinformatics Ma´ıra R. Rodrigues 1⋆ and Michael Luck 1 School of Electronics and Computer Science, University of Southampton, Southampton SO17 1BJ, UK Abstract. In dynamic applications characterised by a variety of alter- native services with the same functionality but heterogeneous results, agents requesting services must find an efficient way to select a service provider from alternatives. In this context, this paper proposes an eval- uation method to analyse the outcome of dynamic service, in order to provide a guide for agents in future decision-making over alternative in- teraction partners. We consider the application of the evaluation method to the bioinformatics domain and present empirical results that support the need for dynamic evaluation of services in that domain. 1 Introduction Bioinformatics is a new field of research characterised by the application of computer technology to the management and analysis of biological data (i.e., to gather, store, analyse and merge genome and protein related information) [1]. Because of the vast quantities of data being generated by several genome and protein sequencing efforts, a large variety of services have been developed to analyse such data. These services are not only heterogeneous in terms of functionalities and results, they are also distributed over the Internet, and in continuous update. Such a dynamic, distributed and heterogeneous environment imposes restrictions on the task of managing and analysing biological data and services, and points to the suitability of an agent-based approach. When an agent is engaged in this kind of environment and needs to delegate to, or request a bioinformatics service from, another agent, it is likely that it will find many alternative agents providing similar services. In essence, there are three ways of selecting a service provider from alterna- tives: by random selection; by identifying the best provider based on the service or provider description given by the providers themselves; or by identifying the provider with the best outcomes over previous interactions. Random selection may allow providers offering poor services to be selected over those offering better services. Selection based on descriptions given by providers does not guarantee that providers are giving correct information or that the described properties are valid when services are performed in different contexts. The more efficient way to select is the third option, by identifying partners with good outcomes ⋆ The first author is supported by Coordena¸c˜ao de Aperfei¸coamento de Pessoal de N´ıvel Superior (CAPES) of the Brazilian Ministry of Education.