Sullivan JJ, Jones AD, Tanji KK
Department of Land, Air, and Water Resources, University of California, Davis 95616, USA.
J Chem Inf Comput Sci. 2000 Sep;40(5):1113-27. doi: 10.1021/ci990222k.
A methodology for the estimation of Hammett substituent constants from computational-based descriptors utilizing quantitative structure activity/property relationships (QSAR/QSPR) formalism is presented. Electronic descriptors derived from quantum chemical calculations and molecular topology were used to generate computational-based analogues of empirical Hammett substituent constants from statistical analysis. Global quantum chemical reaction indices were drawn from frontier orbital theory and density functional theory and formulated from AM1-based calculations. A localized index based on the electrotopological state index was used to encode information on individual group properties. From a training set consisting of 150 meta and para-substituted benzoic acids, statistical analysis of computational-based descriptors as a function of empirical substituent constants yielded a five-parameter QSAR/QSPR model which generates computational-based constants exhibiting a strong correlation with empirical values (r2 = 0.958). Both internal (PRESS) and external (independent testing set of benzoic acids) validation procedures suggest that the electronic effects QSAR/QSPR model derived in this work from computational-based parameters is a statistically viable paradigm. Both predicted and empirical constants were used in Hammett-type validation analyses as functions of chemical, biological, and spectroscopic data for thirty structurally diverse meta and parasubstituted aromatic testing sets. Statistical measures of ensuing correlations were examined and compared, and the empirical and predicted results were of similar quality. Validation results reveal that a large number of computational-based substituent constants can be accurately estimated from semiempirical AM1 frontier orbital energies and electronic structure information obtained directly from substituted benzoic acids without the aid of empirical parametrization.
本文提出了一种利用定量结构活性/性质关系（QSAR/QSPR）形式，从基于计算的描述符估计哈米特取代基常数的方法。从量子化学计算和分子拓扑学中得出的电子描述符，通过统计分析用于生成基于计算的经验哈米特取代基常数类似物。基于前沿轨道理论和密度泛函理论，从基于AM1的计算中得出全局量子化学反应指数。基于电拓扑状态指数的局部指数用于编码关于各个基团性质的信息。从由150种间位和对位取代苯甲酸组成的训练集中，作为经验取代基常数函数的基于计算的描述符的统计分析，产生了一个五参数QSAR/QSPR模型，该模型生成的基于计算的常数与经验值表现出很强的相关性（r2 = 0.958）。内部（PRESS）和外部（苯甲酸独立测试集）验证程序均表明，这项工作中从基于计算的参数得出的电子效应QSAR/QSPR模型是一种在统计上可行的范式。预测常数和经验常数都用于哈米特型验证分析，作为三十个结构多样的间位和对位取代芳香族测试集的化学、生物学和光谱数据的函数。对所得相关性的统计量进行了检查和比较，经验结果和预测结果质量相似。验证结果表明，无需借助经验参数化，就可以从半经验AM1前沿轨道能量和直接从取代苯甲酸获得的电子结构信息中，准确估计大量基于计算的取代基常数。
