Celsie Alena K D, Parnis J Mark
Canadian Environmental Modelling Centre (CEMC), Department of Chemistry, Trent University, 1600 West Bank Road, Peterborough, ON, K9L 0G2, Canada.
Canadian Environmental Modelling Centre (CEMC), Department of Chemistry, Trent University, 1600 West Bank Road, Peterborough, ON, K9L 0G2, Canada.
Chemosphere. 2023 Dec;344:140195. doi: 10.1016/j.chemosphere.2023.140195. Epub 2023 Sep 23.
Henry's law constants (H) for selected probe molecules have been used as descriptors to estimate the COSMO-RS sigma profiles of solvents and solvent mixtures. Henry's law constants were calculated with COSMOtherm for small sets of probe molecules in 155 organic solvents (training set), and these constants subsequently used as descriptors to model the solvent sigma profiles with 61 multiple linear regression (MLR) equations. Subsequent input into COSMOtherm of weighted basis molecule solvent mixtures whose sigma profiles closely matched those modelled for the training set solvents allowed estimation of air-solvent and water-solvent partition ratios for solutes in solvents and solvent mixtures without input of the solvent or solvent mixture identity. The best performing model had 16 descriptors and gave both a training and test set average root-mean square error (RMSE) of 0.008 and an average relative square error (RSE) of 0.07. Partition ratios (K) were then generated for a test set of 251 additional organic solute molecules in solvent/water media where solvents were test set compounds and H constants for the same probe molecules were used as descriptors. The best performing sigma profile model yielded log K RMSE values ranging from 0.17 to 0.92. Finally, this approach was applied to several mixtures ranging from simple binary mixtures to two mixtures considered to be of unknown or variable composition, complex reaction productions or biological materials (UVCBs), namely gasoline and an essential oil mixture. Mixture/water partition ratios were estimated for 251 solutes giving log K RMSE values ranging from 0.24 to 0.88.
已使用选定探针分子的亨利定律常数（H）作为描述符来估算溶剂和溶剂混合物的COSMO-RS σ分布。通过COSMotherm计算了155种有机溶剂（训练集）中少量探针分子的亨利定律常数，随后将这些常数用作描述符，用61个多元线性回归（MLR）方程对溶剂的σ分布进行建模。随后将σ分布与训练集溶剂建模的σ分布紧密匹配的加权基础分子溶剂混合物输入COSMotherm，无需输入溶剂或溶剂混合物的标识，就可以估算溶质在溶剂和溶剂混合物中的气-溶剂和水-溶剂分配比。表现最佳的模型有16个描述符，训练集和测试集的平均均方根误差（RMSE）均为0.008，平均相对平方误差（RSE）为0.07。然后针对另外251种有机溶质分子的测试集生成分配比（K），这些溶质分子存在于溶剂/水介质中，其中溶剂为测试集化合物，并将相同探针分子的H常数用作描述符。表现最佳的σ分布模型产生的log K RMSE值范围为0.17至0.92。最后，将该方法应用于几种混合物，从简单的二元混合物到被认为成分未知或可变的两种混合物、复杂反应产物或生物材料（UVCB），即汽油和精油混合物。估算了251种溶质的混合物/水分配比，得到的log K RMSE值范围为0.24至0.88。
