论文部分内容阅读
我们发展了一种用于预测有机小分子化合物水溶解度(logS)的经验方法XLOGS.它本质上是一种加合模型,采用83种原子/基团类型和3个校正因子作为模型的描述符.该方法还可以根据一个合适的参照分子的logS实验值来计算未知化合物的logS值.我们将XLOGS模型在由4171个化合物组成的训练集上进行了参数化,多元线性回归获得的相关系数(R2)和标准偏差(SD)分别为0.82和0.96单位.将该训练集进一步分为仅含液体化合物和仅含固体化合物的两个子集.XLOGS模型在这两个子集上的回归结果显示前者优于后者(标准偏差分别为0.65单位和0.94单位).还利用log1/S和logP(脂水分配系数)之间的差值来研究XLOGS方法在液体和固体化合物数据集上的表现.研究结果表明:XLOGS等加合法模型更适合应用于这两者差值接近于0的化合物.还将XLOGS和其它三种流行的logS计算模型(包括Qikprop,MOE-logS和ALOGPS)在一个含有132个类药化合物的独立测试集上进行了比较.总体而言,我们的研究结果为加合法模型在水溶解度预测方面的合理应用提供了指导.
We developed an empirical method XLOGS for predicting the water solubility (log S) of small organic compounds, which is essentially an adduct model using 83 atom / radical types and 3 correction factors as model descriptors The method also calculates the log S value of an unknown compound based on the log S experiment of a suitable reference molecule We parameterized the XLOGS model on a training set of 4,171 compounds and the correlation coefficients obtained by multivariate linear regression R2) and standard deviation (SD) were 0.82 and 0.96 units, respectively. The training set was further divided into two subsets containing only liquid compounds and solid compounds only. The regression of the XLOGS model on these two subsets showed that the former (Standard deviations of 0.65 units and 0.94 units, respectively) The difference between log1 / S and logP (liposomal partition coefficient) was also used to study the performance of the XLOGS method on liquid and solid compound datasets.Results Shows that XLOGS and other additive models are more suitable for compounds with the difference between them being close to 0. The XLOGS and three other popular logS calculation models (including Qikprop, MOE-logS and ALOGPS) The results of our study provide guidance for the rational application of the additive model in water solubility prediction.