Integrating multimodal data fusion for comprehensive characterization, antioxidant marker discovery, and geographical...

Platycodonis Radix (PR), a common functional food ingredient, shows quality variation by origin and age, complicating quality control. To address this, we integrated chemical profiling, spectrum-effect analysis for antioxidant marker identification, and multimodal data fusion of hyperspectral imaging (HSI) and MS data to establish a robust evaluation framework. A total of 206 compounds were identified by UPLC-Qtrap-MS/MS and UPLC-Triple-TOF-MS/MS, among which 140 were semi-quantified by scheduled MRM. Antioxidant activities of samples from six regions were assessed by DPPH and ABTS, and spectrum-effect analysis identified 24 key compounds for quantification. Datasets comprising these 24 quantified antioxidant markers, HSI spectra, and image textures were integrated via mid-level data fusion. A stacking ensemble model achieved 93.33% accuracy with precision, recall, and F1 scores above 96%, outperforming single-modality models. This approach offers a robust framework for chemical profiling, functional quality evaluation, and geographical traceability of PR and related food materials.