Effective Search of the Energy Landscape for Protein Folding

We propose a new algorithmic approach for global optimization in protein folding. We use the information found in various local minima to direct the search for the global minimum. In this way, we explore the energy landscape efficiently by considering only the space of local minima instead of the whole feasible space of conformations. Our fundamental approach is to sample only the space of local minima and guide the sampling process by exploring protein structure building blocks found in sampled local minima. These building blocks form the basis of information in searching for the global minimum. In particular we employ an iterative algorithm that begins with an initial pool of local minima; construct a new pool of solutions by combining the various building blocks found in the original pool; take each solution and map them to their representative local minima; and, repeat the process. Our procedure seems to share a great deal of commonality with evolutionary computing techniques. Indeed, we even employ genetic operators in our algorithm. However, unlike existing hybrid evolutionary computing algorithms where local minimization algorithms are simply used to “fine-tune” the solutions, we focus primarily on constructing local minima from previously explored minima and only use genetic operators to assist in diversification. Hence, our total number of iterations/generations were demonstrated (empirically) to be quite low (≈ 50) whereas standard genetic algorithms and Monte Carlo are very high ranging from 150,000 to nearly 20,000,000 generations in order to provide sufficient opportunity for these methods to converge and achieve their best solution. We applied our idea to several proteins from the Protein Data Bank (PDB) using the UNRES model[1]. We compared against Standard Genetic Algorithms(SGA) and Metropolis Monte Carlo(MMC) approaches. In all cases, our new approach computed the lowest energy conformation.