摘要
选取了150种正构烷基酚的热力学性质作为分析对象。为了研究热力学性质的构效关系,引入并简化了主量子拓扑指数0P、1P。考虑到分子对称性对热力学性质的影响,为了方便量化对称程度,将烷基作为类原子处理,定义了烷基的折算点价和烷基相似度的计算方法,借助矩阵形式计算了烷基酚分子对称特征指数Q。将0P、1P、Q与苯环上烷基的数目T、苯环上相邻烷基的对数B作为分子描述符,利用方差膨胀因子VIF进行筛选,确定1P、T、B、Q作为构效关系模型的自变量。采用最佳变量子集回归方法,并根据Akaike信息判据、Kubinyi函数建立了热力学性质定量构效关系模型,所有模型的相关系数均大于99%,甚至达到100%。采用交叉验证方法对模型的稳健性和预测能力进行了检验,结果表明所建模型具有良好的稳健性和预测能力。
The thermodynamic properties of 150 n-alkyl phenols are chosen for investigation. To develop quantitative structure-property relationship models of the thermodynamic properties, topological indexes of principal quantum number 0P and 1P are introduced and simplified. Considering the influence of molecular symmetry on thermodynamic properties, the alkyl is regarded as atom alike to facilitate the symmetry quantization. Based on defining the equivalent valences and similar degrees of alkyls, symmetrical feature index （Q） of n-alkyl phenol is calculated by matrix, which is established with elements of similar degrees. Molecular descriptors, including topological indexes of principal quantum number （0P, 1P）, symmetrical feature index （Q）, number of alkyl （T） and number of pairs of ortho alkyls （B）, are screened with variance inflation factors, which confirms 1P, T, B, and Q as the independent variables in quantitative structure-property relationship models. With leaps and bounds regression method and applying Akaike’s information criterion and Kubinyi function, a series of relationship models between the molecular structures and the thermodynamic properties are constructed, and all the correlation coefficients are larger than 99% even reach 100%. The leave one out cross validation method is adopted to test the stability and the prediction ability of each model, and the result is quite satisfactory. It is demonstrated that the quantitative structure-property relationship models present satisfactory stability and good＆amp;nbsp;prediction ability.
出处
《化工学报》
EI
CAS
CSCD
北大核心
2015年第1期67-78,共12页
CIESC Journal
关键词
污染
热力学性质
计算机模拟
有机化合物
预测
pollution
thermodynamic properties
computer simulation
organic compounds
prediction
