Ren Biye
Research Institute of Materials Science, South China University of Technology, Guangzhou 510640, P. R. China.
J Chem Inf Comput Sci. 2003 Jul-Aug;43(4):1121-31. doi: 10.1021/ci025651o.
Structure-boiling point relationships are studied for a series of oxo organic compounds by means of multiple linear regression (MLR) analysis. Excellent MLR models based on the recently introduced Xu index and the atom-type-based AI indices are obtained for the two subsets containing respectively 77 ethers and 107 carbonyl compounds and a combined set of 184 oxo compounds. The best models are tested using the leave-one-out cross-validation and an external test set, respectively. The MLR model produces a correlation coefficient of r = 0.9977 and a standard error of s = 3.99 degrees C for the training set of 184 compounds, and r(cv) = 0.9974 and s(cv) = 4.16 degrees C for the cross-validation set, and r(pred) = 0.9949 and s(pred) = 4.38 degrees C for the prediction set of 21 compounds. For the two subsets containing respectively 77 ethers and 107 carbonyl compounds, the quality of the models is further improved. The standard errors are reduced to 3.30 and 3.02 degrees C, respectively. Furthermore, the results obtained from this study indicate that the boiling points of the studied oxo compound dominantly depend on molecular size and also depend on individual atom types, especially oxygen heteroatoms in molecules due to strong polar interactions between molecules. These excellent structure-boiling point models not only provide profound insights into the role of structural features in a molecule but also illustrate the usefulness of these indices in QSPR/QSAR modeling of complex compounds.
通过多元线性回归（MLR）分析研究了一系列含氧有机化合物的结构与沸点的关系。基于最近引入的徐指数和基于原子类型的人工智能指数，分别针对包含77种醚和107种羰基化合物的两个子集以及包含184种含氧化合物的组合集，获得了出色的MLR模型。分别使用留一法交叉验证和外部测试集对最佳模型进行了测试。对于184种化合物的训练集，MLR模型的相关系数r = 0.9977，标准误差s = 3.99℃；对于交叉验证集，r（cv）= 0.9974，s（cv）= 4.16℃；对于21种化合物的预测集，r（pred）= 0.9949，s（pred）= 4.38℃。对于分别包含77种醚和107种羰基化合物的两个子集，模型的质量进一步提高。标准误差分别降至3.30和3.02℃。此外，本研究获得的结果表明，所研究的含氧化合物的沸点主要取决于分子大小，还取决于单个原子类型，特别是分子中的氧杂原子，这是由于分子间强烈的极性相互作用所致。这些出色的结构 - 沸点模型不仅为分子中结构特征的作用提供了深刻见解，还说明了这些指数在复杂化合物的QSPR/QSAR建模中的实用性。
