|Title||Data-driven development of liquid chromatography-mass spectrometry methods for combined sample matrices|
|Publication Type||Journal Article|
|Year of Publication||2021|
|Authors||Ge, Z, Zhang, K, Chen, DDY, Yan, B|
|Keywords||Artificial neural network, Data-driven, Herbal medicine formula, Least squares support vector machine, Liquid chromatography-mass spectrometry, MONTE-CARLO SIMULATION|
Herbal medicine formulas (HMFs), the combinations of two or more herbal medicine (HM) ingredients required in a single prescription, are a typical kind of combined sample matrices. LC-MS is a powerful platform for the analyses of such complex samples. The optimization of separation conditions may require a lot of experiments, because multiple analytes need to be separated from a plethora of possible interfering compounds in the sample mixture containing different herbal medicines. To greatly reduce the complexity needed for the optimization of separation conditions, this work proposes a data-driven approach for the systematic development of LC-MS methods for HMFs, using six HMFs created from four HMs (Atractylodis Macrocephalae Rhizoma, Paeoniae Radix Alba, Corydalis Rhizoma and Ophiopogonis Radix) as case-studies. In this approach, the chromatographic peak parameters (like retention times) of the analytes and interfering compounds under different separation conditions were extracted from the LC-MS database of the HMs. Then data-driven models between the chromatographic peak parameters and the separation parameters were built with machine learning methods (r > 0.996 for all the compounds) and used to predict the chromatographic peaks of the analytes and interfering compounds in HMF analyses. Based on the predictions, all of the separation parameters were optimized without any previous experiments on the HMFs. In the validation experiments for the six HMFs, all of the analytes were well separated. The data-driven approach demonstrated enables systematic and rapid development of LC-MS methods for HMFs, and the separation conditions can be efficiently adjusted for different analytes.