Accurate prediction of acidity constants in aqueous solution via density functional theory and self-consistent reaction field methods
Accurate prediction of acidity constants in aqueous solution via density functional theory and self-consistent reaction field methods
复制标题
DOI：
10.1021/jp012533f
复制
发表时间：
2002-02-21
影响因子：
2.9
通讯作者：
Guida, WC
中科院分区：
文献类型：
--
作者：
Klicic, JJ;Friesner, RA;Guida, WC
我们发展了一个通过从头算量子化学和连续溶剂化方法计算有机化合物酸度常数pK（a）的方案。采用大基组的密度泛函（DFT）计算用于确定气相去质子化能量。溶剂化效应处理通过自洽反应场（SCRF）形式主义，涉及精确的数值解的泊松-玻尔兹曼方程。介电半径的参数化为每个感兴趣的官能团，以优化中性和带电物种的溶剂化自由能计算。虽然这些方法的固有精度相当令人印象深刻（误差在几kcal/mol的数量级上），但它还不足以达到我们为pK（a）预测设定的0.5 pK（a）单位的目标精度。因此，通过直接线性拟合到训练集的pK（a）数据，为每个感兴趣的官能团确定两个另外的经验参数，比例因子和加性因子。通过这种额外的参数化，实现了大约0.5 pK（a）单位的平均精度。电离基团的广泛覆盖面，特别侧重于药物活性化合物的化学重要性。除了获得大的和不同的训练集的数据，我们还选择了已知药物的一个子集，其中已经测量了pK（a），并在没有进一步调整参数的情况下对这些化合物进行了预测。尽管这些分子中的许多分子具有相当大的尺寸和复杂性，但结果在质量上与训练集的结果相似，证明了该方法在没有明确参数化的情况下准确处理取代基效应的能力。该方法已从计算的角度进行了优化，使得它是易于处理的，即使是相对较大的药物化合物在50-100个原子的范围内。
We have developed a protocol for computing the acidity constant (pK(a)) of organic compounds via ab initio quantum chemistry and continuum solvation methods. Density functional (DFT) calculations employing large basis sets are used to determine the gas-phase deprotonation energies. Solvation effects are treated via a self-consistent reaction field (SCRF) formalism involving accurate numerical solution of the Poisson-Boltzmann equation. Dielectric radii are parametrized for each functional group of interest to optimize solvation free energy calculations for neutral and charged species. While the intrinsic accuracy of these approaches is quite impressive (errors on the order of a few kcal/mol), it is not quite good enough to achieve the target accuracy that we have set for pK(a) prediction of 0.5 pK(a) units. Consequently, two further empirical parameters, scaling and additive factors, are determined for every functional group of interest by linear fitting directly to pK(a) data for a training set. With this additional parametrization, an average accuracy on the order of 0.5 pK(a) units is achieved. A wide range of coverage of ionizable groups is presented with special focus on chemistry of importance in pharmaceutically active compounds. In addition to obtaining data for large and diverse training sets, we have also selected a subset of known drugs for which pK(a)'s have been measured and made predictions for these compounds without further adjustment of parameters. The results are similar in quality to that of the training set despite the considerable size and complexity of many of these molecules, demonstrating the ability of the method to accurately handle substituent effects without explicit parametrization thereof. The method has been optimized from a computational viewpoint so that it is tractable even for relatively large pharmaceutical compounds in the 50-100 atom range.