Predicting protein structural features with NeuroEvolution of Augmenting Topologies

The study of proteins and the prediction of their three-dimensional structure is one of the most challenging problem in Structural Bioinformatics. Over the last years, several computational strategies have been proposed as a solution to this problem. As revealed by recent CASP experiments, the best results have been achieved by knowledge-based methods. Despite the advances in the development of computational methods, systems and algorithms for solving this complex problem, further research remains to be done. In this paper, we present a new method based on evolving artificial neural networks to extract structural features from experimentally-determined proteins. Structural models can be used to predict the three-dimensional structure of unknown protein sequences and help to reduce the conformational search space of protein molecules. Neural networks using genetic algorithms has shown great promise in this complex learning task. The proposed method has been tested with five protein sequences. The results show that predicted 3-D structures are topologically comparable to their correspondent experimental ones, thus corroborating the effectiveness of our proposal.