An efficient structure learning method in gene prediction

Abstract

This paper proposes an efficient structure learning method to simplify Bayesian network for detecting splice junction site in gene sequences. In this method, nodes in Bayesian networks are selected as features by feature selection algorithm for structure learning. This algorithm is based on genetic algorithm and uses a MAP (maximum a posterior) classifier for this purpose. The result shows that this method can greatly simplify the network while maintains the high accuracy of prediction. The architecture of the optimized network also indicates that the nucleotides close to Donor site are the key elements in the expression of genes.