Um algoritmo genético híbrido aplicado à predição da estrutura de proteínas utilizando o modelo hidrofóbico-polar bidimensional

Abstract

This work suggests the use of an evolutionary computation technique known as genetic algorithms (GAs) for predicting protein structures in the 2D HP model. The methodology has the main proposal the use of an enhanced fitness function, which makes use of the radius of gyration concept. Special genetic operators were developed and added to those commonly used in GAs, besides new strategies to aid the algorithm in the search of protein conformations. These changes led to the development of a user-friendly software system, with several graphical resources and result reports, named GANDALF PRED. A certain amount of experiments were done with the objective of evaluating the influence of GA parameters in the result obtained. Two test cases were set to evaluate the proposed methodology. The first used 9 manually defined chains whose maximum number of hydrophobic non-local bonds is known a priori and length varying from 20 to 85 residues. The results were compared to two other implementations available in the literature. In the second, 7 proteins with globular traits were taken from PDB and translated to the HP model. Their lengths vary from 288 to 842 residues. The results were presented and discussed, since no comparison could be done. For both test cases, the conformations found can be considered good folds.