Multi-Objective Differential Evolution Algorithms for the Protein Structure Prediction Problem
Pedro Henrique Narloch, Mathias J. Krause, Márcio Dorn · 2020
The structural analysis of proteins is an essential step for understanding their biological function. However, the process of the structural determination of these molecules is expensive and time-consuming. In order to reduce these factors, computational methods might be a provocative approach, despite the complexity associated with it. Over the decades, different computational approaches were proposed as well as different energy force fields. As the force fields consider conflicting terms in its composition, multi-objective optimization approaches showed to be suitable to the Protein Structure Prediction problem. In this way, the objective of the current work is to evaluate and compare three multi-objective algorithms, the Non-Dominated Sorting Genetic Algorithm in its second version, the Generalized Differential Evolution in its third version, and the Differential Evolution Multi-Objective. We split the score3 energy function provided by Rosetta into a bi-objective problem. The first objective considers only the non-bonded van der Waals, while the second one is composed of bonded-terms and a secondary structure reinforcement score. Moreover, structural information provided by the Angle Probability List is considered, since this kind of information proved to be reliable in single-objective approaches. Results obtained are analyzed using GDT and RMSD metrics, showing the better capability of Differential Evolution based methods for the problem.