This paper shows how Evolutionary Algorithms (EAs) are used as components in a system for design of protein fingerprints. The system is used for automated mining of data from protein sequence databases, with the purpose of deriving protein family finger-prints. The fingerprints are expressed as patterns, which can be used for recognition of sequences belonging to specific protein families. The system constructs candidate patterns by analyzing multiple sequence alignments, and selecting pattern elements corresponding to evolutionary conserved positions. Since most candidate patterns are too specific, we use stochastic search algorithms for generalization of the candidate patterns. In a previous version of the system a hill-climbing algorithm was used. In this paper we show how results can be substantially improved by using EAs for this task. We also compare a "standard" EA with a host-parasite EA, and show that it can significantly reduce the number of evaluations.
展开▼