A multiobjective approach for protein structure prediction using a steady-state genetic algorithm with phenotypic crowding

In this paper a multiobjective steady-state genetic algorithm with phenotypic crowding is proposed to the (free-modeling) protein structure prediction problem. In this algorithm, the parental replacement is carried out using a criterion associated with the individuals structural similarity, providing a better exploration of highly multimodal energy landscape and the concurrent identification of multiple minima. Classical force field terms, hydrogen bond potentials and a hydrophobic com-pactation term were selected to compose different objectives. Two multiobjective approaches containing two and three objectives were evaluated in predictions performed for a set of 45 proteins. Their results were compared with the ones obtained using a standard single-objective approach. Our results show that the multiobjective approach developed in this work proved to be quite promising in dealing with the PSP problem. In this sense, the algorithm with three objectives showed a better exploration of the search space and obtained significant improvements in the prediction of models for larger proteins and also for structures containing β-sheet secondary structures.