An integrative strategy for effect-linked screening and rapid quantitative prediction of bioactive constituents via multi-source spectral fusion.

Although spectral and chromatographic analysis are broadly applied to compound detection, the quality characterization of complex multicomponent systems remains a persistent challenge. Current analytical paradigms emphasize single-component quantification, lacking compositional coherence in highly coupled chemical systems. Therefore, it is crucial to develop an integrative strategy that enables effect-oriented variable selection and quantitative prediction based on information-rich but low-cost measurements. Using Reduning Injection as a representative case, this study quantitatively characterized 18 components by multi-wavelength fused high-performance liquid chromatography, evaluated antioxidant activity using 2,2-diphenyl-1-picrylhydrazyl free radical scavenging assay, and identified 11 antioxidant activity-associated components based on spectrum-effect relationship analysis. These components were further analyzed by network pharmacology and molecular docking, providing theoretical mapping of biological networks. Prediction models were then developed using ultraviolet, near-infrared, and UV-NIR fusion spectral data for the selected components. Low, Mid, and High-level data-fusion strategies were subsequently employed to optimize the model performance. Among the three UV-NIR fusion levels, all fusion strategies outperformed the single-spectrum models. Mid-level data fusion showed the best overall predictive performance, with MLPRegressor yielding the optimal results among the tested models (mean R2p = 0.951, RMSEp = 0.036, and RPDp = 4.398). These results suggest that the predictive advantage of UV-NIR fusion depended not only on inter-spectral complementarity, but also on effective optimization of the fused feature space. Overall, this study developed an activity-oriented quality evaluation framework that integrates chemical profiling, effect-related component screening, and rapid spectral prediction, providing an integrative approach for the quality characterization and consistency evaluation of TCM preparations.