CE^3: Customizable and Easily Extensible Ensemble Tool for Motif Discovery

Ensemble methods (or simply ensembles) for motif discovery represent a relatively new approach to improve the accuracy of standalone motif finders. In particular, the accuracy of an ensemble is determined by the included finders and the strategy (learning rule) used to combine the results returned by the latter, making these choices crucial for the ensemble success. In this research we propose a general architecture for ensembles, called CE3, which is meant to be extensible and customizable for what concerns external tools inclusion and learning rule. Using CE3 the user will be able to “simulate” existing ensembles and possibly incorporate newly proposed tools (and learning functions) with the aim at improving the ensemble’s prediction accuracy. Preliminary experiments performed with a prototype implementation of CE3 led to interesting insights and a critical analysis of the potentials and limitations of currently available ensembles.