Perfilación sistemática multicomponente de la cápsula Xiangju Rupining mediante cromatografía líquida bidimensional...

Deep learning assisted classification, preferred ion lists guided acquisition, and molecular network visualization analysis (DPM) stepwise acquisition workflows integrate a series of algorithms and functions for the characterization of chemical components. The approach was employed on an online two-dimensional liquid chromatography quadrupole time-of-flight mass spectrometry (2DLC-Q-TOF-MS) analytical platform to systematically characterize the chemical composition of Xiangju Rupining Capsule (XRC). First, an RP-LC × RP-LC system was optimized to maximize orthogonality and peak capacity. Next, an intelligent classification system based on a deep learning mass defect filter (MDF) was developed to generate nine preferred ion lists (PILs) from mass spectral data. A data-dependent (DDA) stepwise acquisition mode was then adopted to iteratively collect PILs, enhancing sensitivity and throughput through dynamic exclusion and background exclusion strategies. In data-independent acquisition (DIA) mode, multi-window acquisition allowed comprehensive coverage of the PIL mass range, ensuring efficient capture of both low- and high-abundance ions. In addition, a multi-source classification-based molecular network was employed to visualize and distinguish different PIL categories. Data annotation was supported by reference standards, characteristic diagnostic ions (CDIs), neutral loss fragments (NLs), as well as relevant literature and multiple online databases. Finally, 191 compounds were tentatively identified in XRC and their potential sources were annotated, including 32 terpenoids, 44 flavonoids, 32 phenolic acids, 37 alkaloids, 8 fatty acids, 10 amino acids, and 28 other types. In summary, the developed workflow provides an efficient method for the acquisition and detailed characterization of complex natural products.