Mining Structural Databases: An Evolutionary Multi-Objetive Conceptual Clustering Methodology

Abstract

The increased availability of biological databases contain ing representations of complex objects permits access to vast amounts of data. In spite of the recent renewed interest in knowledge-discovery tech niques (or data mining), there is a dearth of data analysis methods in tended to facilitate understanding of the represented objects and related systems by their most representative features and those relationship de rived from these features (i.e., structural data). In this paper we propose a conceptual clustering methodology termed EMO-CC for Evolution ary Multi-Objective Conceptual Clustering that uses multi-objective and multi-modal optimization techniques based on Evolutionary Algorithms that uncover representative substructures from structural databases. Be sides, EMO-CC provides annotations of the uncovered substructures, and based on them, applies an unsupervised classification approach to retrieve new members of previously discovered substructures. We apply EMO-CC to the Gene Ontology database to recover interesting sub structures that describes problems from different points of view and use them to explain inmuno-inflammatory responses measured in terms of gene expression profiles derived from the analysis of longitudinal blood expression profiles of human volunteers treated with intravenous endo toxin compared to placebo.