Guvench Olgun, MacKerell Alexander D
Department of Pharmaceutical Sciences, School of Pharmacy, University of Maryland, 20 Penn St., HSF II-629, Baltimore, MD 21201, USA.
J Mol Model. 2008 Aug;14(8):667-79. doi: 10.1007/s00894-008-0305-0. Epub 2008 May 6.
We present a general conformational-energy fitting procedure based on Monte Carlo simulated annealing (MCSA) for application in the development of molecular mechanics force fields. Starting with a target potential energy surface and an unparametrized molecular mechanics potential energy surface, an optimized set of either dihedral or grid-based correction map (CMAP) parameters is produced that minimizes the root mean squared error RMSE between the parametrized and targeted energies. The fitting is done using an MCSA search in parameter space and consistently converges to the same RMSE irrespective of the randomized parameters used to seed the search. Any number of dihedral parameters can be simultaneously parametrized, allowing for fitting to multi-dimensional potential energy scans. Fitting options for dihedral parameters include non-uniform weighting of the target data, constraining multiple optimized parameters to the same value, constraining parameters to be no greater than a user-specified maximum value, including all or only a subset of multiplicities defining the dihedral Fourier series, and optimization of phase angles in addition to force constants. The dihedral parameter fitting algorithm's performance is characterized through multi-dimensional fitting of cyclohexane, tetrahydropyran, and hexopyranose monosaccharide energetics, with the latter case having a 30-dimensional parameter space. The CMAP fitting is applied in the context of polypeptides, and is used to develop a parametrization that simultaneously captures the phi,psi energetics of the alanine dipeptide and the alanine tetrapeptide. Because the dihedral energy term is common to many force fields, we have implemented the dihedral-fitting algorithm in the portable Python scripting language and have made it freely available as "fit_dihedral.py" for download at http://mackerell.umaryland.edu.
我们提出了一种基于蒙特卡罗模拟退火（MCSA）的通用构象能量拟合程序，用于分子力学力场的开发。从目标势能面和未参数化的分子力学势能面开始，生成一组优化的二面角或基于网格的校正图（CMAP）参数，使参数化能量与目标能量之间的均方根误差（RMSE）最小化。拟合是通过在参数空间中进行MCSA搜索完成的，无论用于搜索种子的随机参数如何，都能始终收敛到相同的RMSE。可以同时对任意数量的二面角参数进行参数化，从而实现对多维势能扫描的拟合。二面角参数的拟合选项包括对目标数据进行非均匀加权、将多个优化参数约束为相同值、将参数约束为不大于用户指定的最大值、包括定义二面角傅里叶级数的所有或仅一部分多重性，以及除了力常数之外还优化相角。通过对环己烷、四氢吡喃和己吡喃糖单糖能量学进行多维拟合来表征二面角参数拟合算法的性能，后一种情况具有30维的参数空间。CMAP拟合应用于多肽的情况，并用于开发一种参数化方法，该方法同时捕捉丙氨酸二肽和丙氨酸四肽的φ、ψ能量学。由于二面角能量项在许多力场中都很常见，我们已用便携式Python脚本语言实现了二面角拟合算法，并将其作为“fit_dihedral.py”免费提供，可从http://mackerell.umaryland.edu下载。
