Predicting chromatographic retention times of pesticides has become more and more important for suspect and non-target screening. Indeed, high-resolution mass spectrometry hyphenated (HRMS) to liquid chromatography (LC) are of growing interest for research and monitoring of pesticides, their metabolites and transformation products. The development of quantitative structure-retention relationship models require selecting the most adequate and best set of molecular descriptors and the best machine-learning algorithm. Here, we used seven molecular descriptor sets extracted from four well-known studies and applied them to roughly 800 pesticides and their chromatographic reversed-phase retention times. We used and optimized five different machine-learning algorithms with these descriptor sets to carry out predictions. Our results show that a support-vector machine regression algorithm with only eight molecular descriptors gave the best compromise between the number of molecular descriptors, processing time and model complexity to optimize prediction performance for this specific gradient LC method.

对于可疑和非目标物筛查而言，预测农药的色谱保留时间变得越来越重要。的确，将高分辨率质谱联用（HRMS）联用液相色谱（LC）的研究和监测农药，其代谢物和转化产物的兴趣日益浓厚。定量结构-保留关系模型的发展需要选择最适当和最佳的分子描述符集和最佳的机器学习算法。在这里，我们使用了从四项著名研究中提取的七个分子描述符集，并将它们应用于大约800种农药及其色谱反相保留时间。我们使用这些描述符集使用并优化了五种不同的机器学习算法，以进行预测。