Since the native conformation or the natural shape of a protein largely determines its function, a prediction of protein conformation can shorten the process of drug discovery. This prediction is an optimization search to locate a configuration associated with the global minimum energy for the molecule. Due to the complexity of the multidimensional energy landscape, the prediction process can be extensive, which leads to very long simulation run times. For example, a high-resolution structure prediction algorithm [1] refining 20,000 to 30,000 models of several 49 to 88 residue long molecules takes about 150 CPU days per molecule. This paper presents the method of modified energy landscape (MEL) that improves the efficiency of the Broyden–Fletcher–Goldfarb–Shanno (BFGS) method by 12.8% on average, and more than 30% in some cases for a representative sample of cases. Since the efficiency improvement allows the probabilistic search to cover more areas of the energy landscape, locating the global minimum is more likely. Also, in a practical protein prediction running coarse refinements on more decoys is more preferable than comprehensively refining few decoys because of the low accuracy of energy functions. Therefore, the MEL can significantly improve the protein prediction simulation even though it yields less average score improvement. The MEL is implemented in a refinement protocol in Rosetta [2].

This content is only available via PDF.
You do not currently have access to this content.