Memetic Algorithms for Neural Network Training in Bioinformatics

Bioinformatics is a new, rapidly growing, scientific area that exploits computational techniques to study DNA and protein sequences. A particularly interesting task in this context is to predict the structure of proteins. Artificial Neural Networks can efficiently handle classification and prediction tasks. On the other hand, Memetic Algorithms belong to the class of heuristic methods that have been developed to address hard optimization problems. Since training Artificial Neural Networks is such a problem, Memetic Algorithms can be used to address it. In this paper we propose a LocalSearch-Based Memetic Algorithm, and study its performance as a neural network training method. The resulting Neural Network is applied for the prediction of the cellular localization sites of two proteins. The performance of the proposed method is compared to that of alternative memetic and global optimization algorithms.