Mannhold R, Rekker R F, Sonntag C, ter Laak A M, Dross K, Polymeropoulos E E
Department of Lasermedicine, Heinrich-Hein-Universität, Düsseldorf, Germany.
J Pharm Sci. 1995 Dec;84(12):1410-9. doi: 10.1002/jps.2600841206.
The predictive power of four calculation procedures for molecular lipophilicity is checked by comparing with experimental data (log P and chromatographical RMw) taken from the literature. Two sets of test compounds are used: the first comprises simple organic molecules and the second consists of more complicated drug molecules. Our comparative evaluation leads us to conclude that the predictive power is significantly better for not too complicated organic molecules than for drugs with complicated structural pattern. The four investigated calculation procedures should be arranged in two groups with significantly differing predictive power: (a) Rekker and Hansch/Leo and (b) Ghose/Crippen and Suzuki/Kudo. This conclusion is based on a statistical control using log P and RMw as the independent parameters. Correlations have in common: (1) slopes in correlations with calculated data based on fragmental methods are not significantly different from 1; calculations with data from atom-based procedures show up in most cases with slopes below 1. (2) The accompanying overall statistics underline the superiority of the fragmental methods. We think that all four tested calculation procedures have their own restrictions; for future development we would advise a thorough reconsideration of structural effects not fully (or even not at all) incorporated in the data sets. Special attention will have to be paid to the conformational aspects of lipophilic behavior.
通过与文献中的实验数据（log P和色谱保留值RMw）进行比较，检验了四种分子亲脂性计算方法的预测能力。使用了两组测试化合物：第一组包括简单有机分子，第二组由更复杂的药物分子组成。我们的比较评估得出结论，对于不太复杂的有机分子，预测能力明显优于结构模式复杂的药物。所研究的四种计算方法应分为两组，其预测能力有显著差异：（a）Rekker和Hansch/Leo方法，以及（b）Ghose/Crippen和Suzuki/Kudo方法。这一结论基于以log P和RMw作为独立参数的统计检验。相关性有以下共同特点：（1）基于片段法与计算数据的相关性斜率与1无显著差异；基于原子法的数据计算在大多数情况下斜率低于1。（2） 伴随的总体统计数据突出了片段法的优越性。我们认为所有四种测试计算方法都有其自身的局限性；对于未来的发展，我们建议彻底重新考虑未完全（甚至根本未）纳入数据集的结构效应。必须特别关注亲脂性行为的构象方面。
