Recognition of exon/intron boundaries using dynamic ensembles

Many studies have been carried out in recognition of exon/intron boundaries. For example, PROCRUSTES uses similarity-based approach to gene recognition. Other examples include GRAIL (Gene Recognition and Assembly Internet Link) http://compbio.ornl.gov/Grail-1.3/help/(1996) and Glimmer (Gene Locator and Interpolated Markov Modeler). Since the problem of recognition of exon/intron boundaries can be cast as a classification task, ensemble learning can be applied. An ensemble consists of a set of organized individual trained classifiers whose individual decisions are combined in a certain way for classification purpose. However, existing studies typically take static approaches which hampered flexibility for improved accuracy. To overcome this problem we have proposed the concept of dynamic ensemble and developed a new algorithm, BAGA, which combines bagging and genetic algorithm techniques.