Solving the Protein Structure Prediction Problem with Fast Messy Genetic Algorithms

The ability to accurately predict a polypeptide`s molecular structure given its amino acid sequence is important to numerous scientific, medical, and engineering applications. Studies have been conducted in the application of Genetic Algorithms (GAs) to this problem with promising initial results. In this thesis report, we use the fast messy Genetic Algorithm (fmGA) to attempt to find the minimization of an empirical CHARMM energy model and generation of the associated conformation. Previous work has shown that the fmGA provided favorable results, at least when applied to the pentapeptide [Met]-Enkephalin. We extend these results to a larger Polyalinine14 peptide by utilizing secondary structure information as both searching constraints and seeding the initial population. Additional efforts where conducted to improve the performance of the algorithm with respect to solving the Protein Structure Prediction (PSP) problem through a short-circuiting operator where complete evaluation of the fitness function is halted if initial results are not promising, and by conducting additional searches on faster machines in a heterogeneous environment. Results indicate that, on average, this localized search tends to produce better final solutions. Finally, the fmGA as applied to the PSP problem is analyzed and shown to have improved performance and effectiveness.