白藜芦醇类似物热力学性质的构效关系
QSPR study on thermodynamic properties of resveratrol analogues

摘要： 为了研究热力学性质的构效关系,选取了60种白藜芦醇类似物的热力学性质作为分析对象,引入了主量子拓扑指数{0P2,0P4,0P3'}、{∑0P}。通过对比分析,兼顾模型的简洁性,确定{∑0P}为模型的基本自变量。结合正弦级数的特点,引入{sin(k∑0P)}(k=1,2,3…)到模型中期望消除残差。以调整判定系数最大(R2adj)为目标,在控制变量多重共线性并兼顾变量最少的条件下,建立了各热力学性质构效关系模型。结果表明:引入sin(5∑0P)、sin(8∑0P)到模型中,能较好地消除残差,提高模型的预测能力,所有建立的模型的相关系数均大于0.99,甚至达到1。采用交叉验证方法对模型的稳健性进行了检验,并对样本外分子的热力学性质进行了预测,结果表明所建模型具有良好的稳健性和预测能力。
Abstract: In order to discover the relationships between thermodynamic properties and molecular descriptors, the thermodynamic properties of 60 resveratrol analogues, such as the mean polarizability (α), the standard molar enthalpy of formation(HΘ), the thermal energy(EΘ), the molar heat capacity under constant volume(CΘV)and the standard molar entropy(SΘ), were investigated in this work, and the molecular descriptors, the topological indexes of principal quantum number(0P2,0P4,0P3')and the sum of them(∑0P), were calculated. With less variables, the sum of topological indexes of principal quantum number(∑0P)was confirmed as the base variable for the models through comparative analysis. Base on the characteristics of sine series, the series sin(k∑0P)(where k=1, 2, 3, …)were incorporated into the models to eliminate the residuals. Reducing the multicollinearity among independents and seeking less variables, the quantitative structure-property relationship(QSPR)models of thermodynamic properties were established with the max adjusted coefficient of determination(R2adj)as the destination. The results showed that it was successful to eliminate the residuals and improved the predictabilities by introducing sin(5∑0P)and sin(8∑0P)into the models. And all the correlation coefficients are larger than 0.99, even reach 1. The leave one out cross validation method was adopted to test the stability, and the prediction ability of each model was tested by 7 resveratrol analogues(not in samples), and both results showed quite satisfactory. It has been demonstrated that the quantitative structure-property relationship(QSPR)models have satisfactory stability and good predictability