Puzyn Tomasz, Suzuki Noriyuki, Haranczyk Maciej, Rak Janusz
Research Center for Environmental Risk, National Institute for Environmental Studies, 16-2 Onogawa, Tsukuba, Ibaraki 305-8506 Japan.
J Chem Inf Model. 2008 Jun;48(6):1174-80. doi: 10.1021/ci800021p. Epub 2008 May 30.
Most of the recently published quantitative structure-property relationship (QSPR) models, which can be used to predict environmentally relevant physicochemical data for persistent organic pollutants (e.g., polychlorinated dibenzo- p-dioxins, dibenzofurans, and biphenyls), employ molecular descriptors obtained by means of relatively costly calculations at the density functional theory (DFT) level. However, new semiempirical methods, PM6 and RM1, have recently been developed by J. J. P. Stewart's group. In this study, we compared various QSPR models based on DFT (B3LYP functional) descriptors with the same models based on semiempirical (PM6 and RM1) descriptors. We recalibrated 10 previously published models (for different properties and groups of congeneric compounds) employing PM6 and RM1 descriptors instead of B3LYP ones. We demonstrated that by applying RM1 and PM6 descriptors, we could obtain QSPR models with quality similar to that of models based on B3LYP descriptors. This level of accuracy was out of reach for the models employing AM1- and PM3-based descriptors.
最近发表的大多数定量结构-性质关系（QSPR）模型可用于预测持久性有机污染物（如多氯二苯并-对-二噁英、二苯并呋喃和联苯）与环境相关的物理化学数据，这些模型采用的分子描述符是通过在密度泛函理论（DFT）水平上进行相对昂贵的计算获得的。然而，J. J. P. 斯图尔特团队最近开发了新的半经验方法PM6和RM1。在本研究中，我们将基于DFT（B3LYP泛函）描述符的各种QSPR模型与基于半经验（PM6和RM1）描述符的相同模型进行了比较。我们重新校准了10个先前发表的模型（针对不同性质和同系物组化合物），使用PM6和RM1描述符而非B3LYP描述符。我们证明，通过应用RM1和PM6描述符，我们可以获得质量与基于B3LYP描述符的模型相似的QSPR模型。这种准确度水平是采用基于AM1和PM3描述符的模型所无法达到的。
