Classifying Patterns in Bioinformatics Databases by Using Alpha-Beta Associative Memories

Abstract

One of the most important genomic tasks is the identification of promoters and splice-junction zone, which are essential on deciding whether there is a gene or not in a genome sequence. This problem could be seen as a classification problem, therefore the use of computational algorithms for both, pattern recognition and classification are a natural option to face it. In this chapter we develop a pattern classifier algorithm that works notably with bioinformatics databases. The associative memories model on which the classifier is based is the Alpha-Beta model. In order to achieve a good classification performance it was necessary to develop a new heteroassociative memories algorithm that let us recall the complete fundamental set. The heteroassociative memories property of recalling all the fundamental patterns is not so common; actually, no previous model of heteroassociative memory can guarantee this property. Thus, creating such a model is an important contribution. In addition, an heteroasociative Alpha-Beta multimemory is created, as a fundamental base for the proposed classifier.