Ab initio protein folding simulations with genetic algorithms: Simulations on the complete sequence of small proteins

Abstract

Ab-initio folding simulations have been performed on three small proteins using a genetic algorithm- (GA-) based search method which operates on an all atom representation. Simulations were also performed on a number of small peptides expected to be independent folding units. The present genetic algorithm incorporates the results of developments made to the method first tested in CASP1. Additional operators have been introduced into the search in order to allow the simulation of longer sequences and to avoid the simulation of longer sequences and to avoid premature free energy convergence. Secondary structure information derived from a consensus of eight methods and Monte Carlo simulations on sets of homologous sequences has been used to bias the starting populations used in the GA simulations. For the fragment simulations, the results generally have approximately correct local structure, but tend to be too compact, leading to poor RMS error values. One of the three small protein structures has the topology and most of the general organization correct, although many of the details are incorrect.