RICHARD O. DAY and GARY B. LAMONT
Interest in discerning protein structures through prediction (PSP) is widespread and has been previously addressed using a variety of techniques including X-ray crystallography, molecular dynamics, nuclear magnetic resonance spectroscopy, Monte Carlo analysis, lattice simulation, and evolutionary algorithms (EAs) in energy minimization. We have employed EAs such as the simple genetic algorithm (GA), messy GA (mGA), and the linkage learning GA (LLGA), with associated public domain software. Here, we report results of a parallel modified fast messy GA (fmGA), which is found to be quite "good" at finding semi-optimal PSP solutions in a reasonable time. We focus on modifications to this EA called the fmGA, extensions to the multiobjective implementation of the fmGA (MOfmGA), constraint satisfaction via Ramachandran plots, identifying secondary protein structures, a farming model for the parallel fmGA (pfmGA), and fitness function approximation techniques. These techniques reflect marked improvement over previous GA applications for protein structure determination. Problem definition, protein model representation, mapping to algorithm domain, tool selection modifications, and conducted experiments are discussed.
Was this article helpful?