Ajmani Subhash, Rogers Stephen C, Barley Mark H, Livingstone David J
Centre for Molecular Design, Institute of Biomedical and Biomolecular Science, University of Portsmouth, King Henry 1 Street, Portsmouth PO1 2DY, UK.
J Chem Inf Model. 2006 Sep-Oct;46(5):2043-55. doi: 10.1021/ci050559o.
In this paper we report an attempt to apply the QSPR approach for the analysis of data for mixtures. This is an extension of the conventional QSPR approach to the analysis of data for single molecules. The QSPR methodology was applied to a data set of experimental measured density of binary liquid mixtures compiled from the literature. The present study is aimed to develop models to predict the "delta" value of a mixture i.e., deviation of the experimental mixture density (MED) from the ideal, mole-weighted calculated mixture density (MCD). The QSPR was investigated in two perspectives (QMD-I and QMD-II) with respect to the creation of training and test sets. The study resulted in significant ensemble neural network and k-nearest neighbor models having statistical parameters r2, q2(10cv) greater than 0.9, and pred_r2 greater than 0.75. The developed models can be used to predict the delta and hence the density of a new mixture. The QSPR analysis shows the importance of hydrogen bond, polar, shape, and thermodynamic descriptors in determining mixture density, thus aiding in the understanding of molecular interactions important in molecular packing in the mixtures.
在本文中，我们报告了将定量结构-性质关系（QSPR）方法应用于混合物数据分析的尝试。这是将传统QSPR方法扩展到单分子数据分析。QSPR方法应用于从文献中汇编的二元液体混合物实验测量密度的数据集。本研究旨在开发模型以预测混合物的“δ”值，即实验混合物密度（MED）与理想的、摩尔加权计算的混合物密度（MCD）的偏差。关于训练集和测试集的创建，从两个角度（QMD-I和QMD-II）对QSPR进行了研究。研究得到了具有统计参数r2、q2(10cv)大于0.9且pred_r2大于0.75的显著的集成神经网络和k近邻模型。所开发的模型可用于预测δ，进而预测新混合物的密度。QSPR分析表明氢键、极性、形状和热力学描述符在确定混合物密度方面的重要性，从而有助于理解混合物中分子堆积中重要的分子间相互作用。
