Abstract

The Protein Structure Prediction or 3D-PSP is a relevant problem in bioinformatics. Its computational complexity classifies it as an NP-Hard problem. The 3D-PSP can be approached as a multi-objective optimisation problem by using different energy functions as objective functions and analyzing their conflicting roles in the protein structure conformation. This work examines the energy functions Talaris2013 and SASA of the molecular modelling suite PyRosetta, using a version of the multi-objective NSGA-II algorithm, plus a local search heuristics and the use of the Angle Probability List to incorporate structural information in the optimisation process. The experiment results showed the different aspects each energy function evaluates and the better structure prediction when used together in a multi-objective approach than separately in the optimisation process. The results also showed improved RMSD and GDT values compared to the equivalent results of previous works. Also, the results showed better RMSD values with the Pareto Local Search algorithm version than without a local search.