Prediction of properties from simulations: Free energies of solvation in hexadecane, octanol, and water
Prediction of properties from simulations: Free energies of solvation in hexadecane, octanol, and water
复制标题
DOI：
10.1021/ja993663t
复制
发表时间：
2000-03-29
影响因子：
15
通讯作者：
Jorgensen, WL
中科院分区：
文献类型：
--
作者：
Duffy, EM;Jorgensen, WL
对水溶液中的 200 多种有机溶质（包括 125 种药物和相关杂环化合物）进行了蒙特卡罗 (MC) 统计力学模拟。计算高度自动化，并使用由 CM1P 部分电荷增强的 OPLS-AA 力场。获得了各种重要物理量的构型平均结果，包括溶质-水库仑和伦纳德-琼斯相互作用能、溶剂可及表面积 (SASA) 以及供体和受体氢键的数量。然后获得这些描述符与十六烷、辛醇和水的溶剂化气液自由能以及辛醇/水分配系数之间的相关性。在所有情况下，具有三个或四个描述符的线性回归均产生相关系数 r(2) 为 0.9 的拟合。 log P（辛醇/水）的回归方程仅需要四个描述符即可为 200 种不同化合物提供 0.55 的均方根误差，这与最佳片段方法具有竞争力。对于水，85 种溶质的扩展数据集和改进的统计分析使人们对先前线性响应处理中的 Lennard-Jones 和表面积项的重要性产生了疑问。结果对溶质原子部分电荷的选择很敏感；某些官能团的代表性较差可能导致需要对回归方程进行特定修正。预计对于蛋白质-配体结合的基于力场的评分函数也是如此。在所有情况下，目前最重要的描述符揭示了控制溶剂化的关键物理因素，特别是有机溶剂中的溶质尺寸和水中的静电相互作用。此外，对水和乙醇中溶质的额外 MC 模拟清楚地表明，水和醇之间的关键区别在于水具有更强的氢键供给能力，这解释了溶质接受氢键的重要性； log P（辛醇/水）的能力。
Monte Carlo (MC) statistical mechanics simulations have been carried out for more than 200 organic solutes, including 125 drugs and related heterocycles, in aqueous solution. The calculations were highly automated and used the OPLS-AA force field augmented with CM1P partial charges. Configurationally averaged results were obtained for a variety of physically significant quantities including the solute-water Coulomb and Lennard-Jones interaction energies, solvent-accessible surface area (SASA), and numbers of donor and acceptor hydrogen bonds. Correlations were then obtained between these descriptors and gas to liquid free energies of solvation in hexadecane, octanol, and water and octanol/water partition coefficients. Linear regressions with three or four descriptors yielded fits with correlation coefficients, r(2), of 0.9 in all cases. The regression equation for log P(octanol/water) only needs four descriptors to provide an rms error of 0.55 for 200 diverse compounds, which is competitive with the best fragment methods. For water, the expanded data set of 85 solutes and improved statistical analyses bring into question the significance of the Lennard-Jones and surface area terms that have been featured in prior linear-response treatments. The results are sensitive to the choice of partial charges for the solute atoms; poor representation of some functional groups can lead to the need for specific corrections in the regression equations. This is expected to also be true for force-field-based scoring functions for protein-ligand binding. In all cases, the present descriptors that emerge as most significant sensibly reveal the key physical factors that control solvation, especially solute size in organic solvents and electrostatic interactions in water. Furthermore, additional MC simulations for solutes in both water and ethanol clearly demonstrate that the key differential between water and alcohols is the greater hydrogren-bond-donating ability of water, which explains the significance of a solute's hydrogen-bond-accepting; ability for log P(octanol/water).