D'Archivio Angelo Antonio, Maggi Maria Anna, Mazzeo Pietro, Ruggieri Fabrizio
Dipartimento di Chimica, Ingegneria Chimica e Materiali, Università degli Studi di L'Aquila, Via Vetoio, 67010 Coppito, L'Aquila, Italy.
Anal Chim Acta. 2008 Nov 3;628(2):162-72. doi: 10.1016/j.aca.2008.09.018. Epub 2008 Sep 12.
The ability of the Weighted Holistic Invariant Molecular (WHIM) and GEometry, Topology, and Atom-Weights AssemblY (GETAWAY) descriptors to represent the effect of molecular structure on the retention of pesticides in reversed-phase high-performance liquid chromatography (RP-HPLC) is investigated. To this end, two retention data sets previously collected in water-acetonitrile mobile phase are re-examined. The first data set (data-set-1) consists of retention data of 26 neutral compounds belonging to widely used pesticide classes, collected within the mobile phase composition range 40-65% (v/v) acetonitrile. The second data set (data-set-2) describes retention of phenoxy acid herbicides and structurally related compounds (benzoic acid and phenylacetic acid derivatives), as a whole covering the pK(a) range 2.3-4.3, as a function of mobile phase composition, ranging between 30 and 70% (v/v) acetonitrile, and pH, ranging between 2 and 5. For each data set, the mobile phase attributes are combined with WHIM or GETAWAY descriptors into "mixed" predictive models in order to attempt retention modelling within the whole mobile phase composition range of analytical interest. Six- or seven-dimensional multilinear models, preliminarily selected using a genetic algorithm, were improved using a multi-layer artificial neural network (ANN) learned by back propagation. ANN performance was tested on three molecules not used in the learning stage and by leave-more-out cross validation. The results reveal that while WHIM descriptors seem not adequate to model retention of solutes of data-set-1, GETAWAY descriptors provide a satisfactory retention model. On the other hand WHIM and GETAWAY descriptors applied to data-set-2 provide a similar performance, even if slightly worse as compared with the above case. Accuracy of retention modelling in these cases is comparable or slightly poorer as compared with the results previously obtained by combining quantum chemical descriptors or usual physico-chemical solute properties (log k(ow) and pK(a)) and mobile phase attributes.
研究了加权整体不变分子（WHIM）描述符和几何、拓扑与原子权重组合（GETAWAY）描述符在反相高效液相色谱（RP-HPLC）中表征分子结构对农药保留的影响的能力。为此，重新审视了之前在水-乙腈流动相中收集的两个保留数据集。第一个数据集（数据集1）由26种属于广泛使用的农药类别的中性化合物的保留数据组成，这些数据是在乙腈体积分数为40-65%（v/v）的流动相组成范围内收集的。第二个数据集（数据集2）描述了苯氧基酸除草剂及结构相关化合物（苯甲酸和苯乙酸衍生物）的保留情况，整体涵盖了2.3-4.3的pKa范围，作为流动相组成（乙腈体积分数在30%至70%（v/v）之间）和pH值（在2至5之间）的函数。对于每个数据集，将流动相属性与WHIM或GETAWAY描述符组合成“混合”预测模型，以便在分析感兴趣的整个流动相组成范围内尝试进行保留建模。使用遗传算法初步选择的六维或七维多线性模型，通过反向传播学习的多层人工神经网络（ANN）进行了改进。ANN性能在学习阶段未使用的三个分子上进行了测试，并通过留多法交叉验证进行了检验。结果表明，虽然WHIM描述符似乎不足以对数据集1中溶质的保留进行建模，但GETAWAY描述符提供了一个令人满意的保留模型。另一方面，应用于数据集2的WHIM和GETAWAY描述符提供了相似的性能，尽管与上述情况相比略差。与之前通过组合量子化学描述符或常用的物理化学溶质性质（log k(ow)和pKa）以及流动相属性获得的结果相比，这些情况下保留建模的准确性相当或略差。
