The search and identification of organic contaminants in agricultural watersheds has become a crucial effort to better characterize watershed contamination by pesticides. The past decade has brought a more holistic view of watershed contamination via the deployment of powerful analytical strategies such as non-target and suspect screening analysis that can search more contaminants and their transformation products. However, suspect screening analysis remains broadly confined to known molecules, primarily due to the lack of analytical standards and suspect databases for unknowns such as pesticide transformation products. Here we developed a novel workflow by cross-comparing the results of various in silico prediction tools against literature data to create an enhanced database for suspect screening of pesticide transformation products. This workflow was applied on tebuconazole, used here as a model pesticide, and resulted in a suspect screening database counting 291 transformation products. The chromatographic retention times and tandem mass spectra were predicted for each of these compounds using 6 models based on multilinear regression and more complex machine-learning algorithms. This comprehensive approach to the investigation and identification of tebuconazole transformation products was retrospectively applied on environmental samples and found 6 transformation products identified for the first time in river water samples.

农业流域中有机污染物的搜索和鉴定已成为更好地表征流域农药污染的关键工作。在过去的十年中，通过部署强大的分析策略（例如可以搜索更多污染物及其转化产物的非目标和可疑筛选分析），对流域污染有了更全面的认识。然而，可疑筛选分析仍然广泛局限于已知分子，这主要是由于缺乏分析标准和未知物（如农药转化产物）的可疑数据库。在这里，我们通过交叉比较各种in silico的结果开发了一种新颖的工作流程针对文献数据的预测工具，以创建用于农药转化产物可疑筛选的增强型数据库。该工作流程应用于戊唑醇（此处用作农药模型），并生成了一个包含 291 种转化产物的可疑筛选数据库。使用基于多线性回归和更复杂的机器学习算法的 6 个模型预测了每种化合物的色谱保留时间和串联质谱。这种全面的戊唑醇转化产物调查和鉴定方法回顾性地应用于环境样品，在河水样品中首次鉴定出6种转化产物。