Structure prediction is often modelled as an optimization problem in which a computer must minimize the potential energy of a molecular system in terms of its atomic coordinates. Potential energy surfaces for chemical systems present a large number of local minima which frustrate the search for global minima. This roughness of conformational space is the primary impediment to conformational search methods. Most present methods for optimization and conformational search do not address the roughness issue directly, but instead rely on stochastic or heuristic mechanisms to traverse surface barriers.
The method investigated herein mathematically and analytically transforms an original potential energy surface into a continuous series of progressively smoother surfaces. Among the achievements of this research are a deformable variant of the AMBER/OPLS potential function and its application to conformational search problems. A thorough analysis of potential smoothing applied to capped dialanine peptide reveals that the deformed surfaces retain the most dominant features of the original surface but have many fewer minima. Consequently, deformed surfaces are easier to search. Strong qualitative and quantitative correlations are identified between potential smoothing and simulated annealing, the present state-of-the-art technique for conformational optimization. These observations lead to the conclusion that potential smoothing is a deterministic analog of simulated annealing.
The analysis of potential smoothing applied to capped dialanine peptide identifies three characteristics of potential smoothing: merging, shifting, and crossing. Each of these effects has a direct correlate in simulated annealing. Crossing is shown to account for the reduced efficacy of potential smoothing and simulated annealing for the global optimization of chemical systems. Analysis of this feature of potential smoothing led to the coupling of potential smoothing to a local search procedure which attempts to correct for crossing events. The results of the hybrid Potential Smoothing and Search (PSS) method applied cycloheptadecane and molecular docking of trypsin-benzamidine and HIV protease-XK263 are presented. Refinements to the concepts presented herein will enable investigations of greater computational complexity than currently possible.