Identifying isomeric metabolites remains a challenging and time-consuming process with both sensitivity and unambiguous structural assignment typically only achieved through the combined use of LC-MS and NMR. Ion mobility mass spectrometry (IMMS) has the potential to produce timely and accurate data using a single technique to identify drug metabolites, including isomers, without the requirement for in-depth interpretation (cf. MS/MS data) using an automated computational pipeline by comparison of experimental collision cross-section (CCS) values with predicted CCS values. An ion mobility enabled Q-Tof mass spectrometer was used to determine the CCS values of 28 (14 isomeric pairs of) small molecule glucuronide metabolites, which were then compared to two different in silico models; a quantum mechanics (QM) and a machine learning (ML) approach to test these approaches. The difference between CCS values within isomer pairs was also assessed to evaluate if the difference was large enough for unambiguous structural identification through in silico prediction. A good correlation was found between both the QM- and ML-based models and experimentally determined CCS values. The predicted CCS values were found to be similar between ML and QM in silico methods, with the QM model more accurately describing the difference in CCS values between isomer pairs. Of the 14 isomeric pairs, only one (naringenin glucuronides) gave a sufficient difference in CCS values for the QM model to distinguish between the isomers with some level of confidence, with the ML model unable to confidently distinguish the studied isomer pairs. An evaluation of analyte structures was also undertaken to explore any trends or anomalies within the data set.

中文翻译：

---

使用硅胶和实验碰撞横截面值研究小分子异构葡萄糖醛酸代谢物分化。

鉴定异构代谢物仍然是一个具有挑战性和耗时的过程，其灵敏度和明确的结构分配通常只能通过组合使用 LC-MS 和 NMR 来实现。离子淌度质谱 (IMMS) 有可能使用单一技术产生及时和准确的数据来识别药物代谢物，包括异构体，而无需使用自动化计算管道进行深入解释（参见 MS/MS 数据）实验碰撞截面 (CCS) 值与预测的 CCS 值的比较。启用离子迁移率的 Q-Tof 质谱仪用于确定 28 种（14 对异构体）小分子葡萄糖醛酸代谢物的 CCS 值，然后将其与两种不同的计算机模型进行比较；量子力学 (QM) 和机器学习 (ML) 方法来测试这些方法。还评估了异构体对内 CCS 值之间的差异，以评估差异是否足够大以通过计算机预测进行明确的结构识别。在基于 QM 和基于 ML 的模型与实验确定的 CCS 值之间发现了良好的相关性。发现 ML 和 QM in silico 方法的预测 CCS 值相似，QM 模型更准确地描述了异构体对之间 CCS 值的差异。在 14 个异构体对中，只有一个（柚皮素葡萄糖醛酸苷）在 QM 模型的 CCS 值上有足够的差异，以有一定的置信度区分异构体，而 ML 模型无法自信地区分所研究的异构体对。