A comprehensive strategy integrating chemical profiling with deep learning for screening the antithrombin bioactive-chemical quality markers of functional food: Pueraria radix as an example
Meiqi Liu, Zicheng Ma, Yalan Li, Piaopiao Yang, Meng Wang, Yanan Liu, Lili Sun, Meiling Chen, Xiaoliang Ren
Abstract
Pueraria lobata radix (PL) and Puerariae tbatae radix (PT) are nutritious medicinal food homology plants that are widely used in the food and health products industry. Both of them are used as Pueraria radix (PR) for medicine. In this study, the function-related active markers were discovered by chemical profiling coupled with deep learning for assessment of PR. Firstly, IR spectral fingerprints were established. And the IR spectra of PL and PT were different in fingerprint region. HCA, PCA and PLS-DA can distinguish the samples of PL and PT. Furtherly, UPLC-Q-TOF-MS was further applied to identify the main components of PL and PT, and 30 chemical components in PL and 15 in PT were identified. In vitro antithrombin assay was performed to indicate the different batches of PL and PT extracts and they could be distinguished according to the strength of activity. The potential active markers related to antithrombin activity were screened by PCA and PLS-DA. Finally, 3'-hydroxy puerarin, puerarin, 3'-methoxy puerarin, daidzin, and daidzein were screened-out as functional markers applying molecular docking combined with BP-ANN, CP-ANN, SIMCA, and SVMDA. The thrombin-based discovery strategy of bioactive-chemical marker combination was a powerful tool for screening the quality markers for evaluation of PR. • It is feasible to estimate Pueraria radix quality based on the UPLC-Q-TOF-MS and MIR spectra. • There were significant differences in antithrombin activity between PL and PT. • Four deep learning models were used for validation. • Five compounds with antithrombin activity were identified in Pueraria radix. • The function-related active markers were discovered for assessment of functional food.