Protein Structure Optimization using a Combinatorial Search Algorithm

We developed a combinatorial search algorithm which we call best profile search for the global optimization of protein structures. This algorithm finds near optimal solutions in an early stage of the optimization. We developed a protein structure scoring function which depends on the distances and orientations of amino acid pairs. We devised an iterative procedure in order to improve the predictive power of the scoring function: We use this best profile search optimization procedure for the ab initio computation of low energy structures of a test set of proteins using our scoring function. This set of ab initio structure calculation is embedded into a second level of optimization: The mean RMSD results of the structure optimizations are taken as an objective function for a simulated annealing procedure. The simulated annealing algorithm minimizes the geometric mean RMSD deviations of the computed structures with respect to their native structures. RESULTS: With the developed scoring function we obtain for proteins of the test set for ab initio structure optimizations a RMSD deviation of 5.2 Angstroem in the geometric mean, taking the best structure among the top ten scoring structures found during structure optimization