Hopkins Chad W, Roitberg Adrian E
Department of Physics and ‡Department of Chemistry, Quantum Theory Project, University of Florida , Gainesville, Florida 32611, United States.
J Chem Inf Model. 2014 Jul 28;54(7):1978-86. doi: 10.1021/ci500112w. Epub 2014 Jul 9.
The derivation and optimization of most energy terms in modern force fields are aided by automated computational tools. It is therefore important to have algorithms to rapidly and precisely train large numbers of interconnected parameters to allow investigators to make better decisions about the content of molecular models. In particular, the traditional approach to deriving dihedral parameters has been a least-squares fit to target conformational energies through variational optimization strategies. We present a computational approach for simultaneously fitting force field dihedral amplitudes and phase constants which is analytic within the scope of the data set. This approach completes the optimal molecular mechanics representation of a quantum mechanical potential energy surface in a single linear least-squares fit by recasting the dihedral potential into a linear function in the parameters. We compare the resulting method to a genetic algorithm in terms of computational time and quality of fit for two simple molecules. As suggested in previous studies, arbitrary dihedral phases are only necessary when modeling chiral molecules, which include more than half of drugs currently in use, so we also examined a dihedral parametrization case for the drug amoxicillin and one of its stereoisomers where the target dihedral includes a chiral center. Asymmetric dihedral phases are needed in these types of cases to properly represent the quantum mechanical energy surface and to differentiate between stereoisomers about the chiral center.
现代力场中大多数能量项的推导和优化借助于自动化计算工具。因此，拥有能快速且精确地训练大量相互关联参数的算法很重要，这样研究人员就能对分子模型的内容做出更好的决策。特别是，传统的二面角参数推导方法是通过变分优化策略对目标构象能量进行最小二乘拟合。我们提出一种计算方法，用于同时拟合力场二面角振幅和相位常数，该方法在数据集范围内是解析的。这种方法通过将二面角势重铸为参数的线性函数，在一次线性最小二乘拟合中完成量子力学势能面的最优分子力学表示。我们就计算时间和两个简单分子的拟合质量，将所得方法与遗传算法进行了比较。正如先前研究中所表明的，只有在对手性分子建模时才需要任意的二面角相位，目前使用的药物中有一半以上是手性分子，所以我们还研究了药物阿莫西林及其一种立体异构体的二面角参数化情况，其中目标二面角包含一个手性中心。在这类情况下需要不对称的二面角相位来正确表示量子力学能量面，并区分手性中心周围的立体异构体。
