DEEPSAM: A Hybrid Evolutionary Algorithm for the Prediction of Biomolecules Structure

摘要

DEEPSAM (Diffusion Equation Evolutionary Programming Simulated Annealing Method), a hybrid evolutionary algorithm, is presented here. This algorithm has been designed for finding the global minimum, and other low-lying minima, of the potential energy surface (PES) of biological molecules. It hybridizes Evolutionary Programming (EP) with two well-known global optimization methods (the Diffusion Equation Method - DEM, and a kind of Simulated Annealing - SA), and with the L-BFGS quasi-Newton local minimization procedure. This combination has produced a powerful tool (a) for finding a good approximation of the native structure of a protein or peptide, given a Force Field (FF) parameters set and a starting (unfolded) structure, and (b) for finding an ensemble of structures close enough structurally and energetically to the native structure. The results obtained until now show that DEEPSAM is a powerful structure predictor, when a reliable FF parameters set is available. DEEPSAM's implementation is time-efficient, and requires modest computational resources.