Parameterizing genetic algorithms for protein folding simulation

A genetic algorithm is used to search energetically and structurally favorable conformations. The authors use a hybrid protein representation, three operators to manipulate the protein `genes´, and a fitness function based on a simple force field. The prototype was applied to the ab initio prediciton of Crambin. None of the conformations generated with a non-biased fitness function are similar to the native conformation but all of them show a much better overall fitness than the native structure. If guided by r.m.s. deviation the native conformation was reproduced at 1.3 Å. Therefore, the genetic algorithm´s search was successful but the fitness function was no good indicator for native structure. In a side chain placement experiment Crambin was reproduced at 1.86 Å r.m.s deviation