Department of Chemistry, Rm 185 Chemistry, Wayne State University, Detroit, MI 48202, USA.
J Chromatogr A. 2020 Oct 11;1629:461500. doi: 10.1016/j.chroma.2020.461500. Epub 2020 Aug 19.
To facilitate faster selectivity evaluation of wall-coated, open-tubular columns using the solvation parameter model a reduced set of calibration compounds is identified and validated for the temperature ranges 60-140 °C and 160-260 °C. The Kennard-Stone uniform mapping algorithm is used to identify the calibration compounds from a larger database of compounds with known retention properties previously adopted for column selectivity evaluation. Thirty-five compounds for each temperature range are required to minimize the standard deviation of the system constants used for selectivity evaluation and to minimize differences between system constants determined by conventional calibration and the reduced calibration compounds. The models for the reduced calibration compounds on ten siloxane-based and poly(ethylene glycol) stationary phases have a coefficient of determination of 0.984 to 0.998 and standard error of the estimate of 0.012 to 0.30. The predictive capability of models is evaluated for the reduced sets of calibration compounds using external test sets with ranking of the calibration models by changes in the average error, average absolute error and root mean square error of prediction for the test sets. For the selected thirty-five reduced calibration compounds the range for the average absolute error was 0.014 to 0.033 and 0.016 to 0.040 for the root mean square error of prediction for the independent test sets.
为了使用溶剂化参数模型更快速地评估壁涂开管柱的选择性，确定并验证了一组用于 60-140°C 和 160-260°C 温度范围的减少的校准化合物。肯纳德-斯通均匀映射算法用于从先前用于柱选择性评估的具有已知保留性质的更大化合物数据库中识别校准化合物。每个温度范围需要 35 种化合物，以最小化用于选择性评估的系统常数的标准偏差，并最小化通过常规校准和减少的校准化合物确定的系统常数之间的差异。十个硅氧烷和聚（乙二醇）固定相上减少的校准化合物模型的决定系数为 0.984 至 0.998，估计的标准误差为 0.012 至 0.30。使用外部测试集通过平均误差、预测测试集的平均绝对误差和均方根误差的变化来对校准模型进行排名，评估了减少的校准化合物集的模型预测能力。对于选定的 35 种减少的校准化合物，平均绝对误差的范围为 0.014 至 0.033，预测独立测试集的均方根误差的范围为 0.016 至 0.040。
