Yang Liu, Man Yang, Han Qin, Jiake Wen, Peng Liu, Xiaoyan Wang, Kunze Du, Yanxu Chang
Comprehensive characterization of the complex chemical constituents in traditional Chinese medicine (TCM) formulas remains a major challenge in elucidating their pharmacodynamic material basis and mechanisms of action. Although liquid chromatography-mass spectrometry (LC-MS) is widely used for constituent identification, conventional acquisition strategies often fail to obtain sufficient fragment-ion information for both high- and low-abundance compounds co-eluting within the same retention-time window. To address this limitation, an online comprehensive two-dimensional liquid chromatography-quadrupole time-of-flight tandem mass spectrometry (2DLC-Q-TOF-MS/MS) platform was integrated with a dual-mode acquisition strategy driven by deep learning-based clustering. Deep learning-assisted mass defect filtering (MDF) was employed to classify precursor ions into specific chemical categories. The Python-based PyMassTime-Cluster system was then used to redistribute highly overlapping, co-eluting predicted ions into balanced sub-priority ion capture (sub-PIC) lists. These sub-PIC lists guided stepwise data-dependent acquisition (DDA), while data-independent acquisition (DIA) was incorporated as a complementary mode to improve overall fragment-ion coverage. This integrated workflow enabled efficient acquisition of MS/MS data for both high- and low-abundance constituents. Using this strategy, 181 compounds were characterized in Qishen Yiqi Dropping Pills (QSYQDP), including 49 flavonoids and their glycosides, 40 organic acids, 27 terpenoids, 12 phenylpropanoids, 10 saponins, 9 amino acids, and 34 other compounds. Overall, the integration of deep learning-based classification, retention-time clustering, and complementary acquisition modes provides an effective and transferable approach for the comprehensive chemical characterization of complex TCM formulas.